609e1e9763
9860 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
ryan-crabbe-berri
|
1cce49b9d0
|
fix(vector-stores): support engines URL for Vertex AI Search (#27885)
Adds optional vertex_engine_id field to vertex_ai/search_api so users can route through a Discovery Engine search app instead of the data store directly. Required for website, healthcare, and connector-based data stores that return FAILED_PRECONDITION on the existing dataStores URL. Existing data-store-direct callers are unaffected. Resolves LIT-3036 |
||
|
Sameer Kankute
|
c908505e6a
|
fix(proxy): omit OpenAI [DONE] on google-genai streamGenerateContent (#29426)
* fix(proxy): omit OpenAI [DONE] on google-genai streamGenerateContent google-genai SDK uses ?alt=sse and cannot parse the proxy's trailing data: [DONE] chunk. Skip that terminator for agenerate_content_stream. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(proxy): address Greptile review on google-genai stream fix Always yield stream error_message; only gate data: [DONE] on the skip flag. Set _litellm_skip_openai_stream_done in google_endpoints instead of common_request_processing. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> |
||
|
milan-berri
|
29270a36a5
|
fix(anthropic, fireworks): inline legacy $ref defs in tool schemas (#28646)
Tools sourced from MCP servers and OpenAPI-derived gateways (AWS
AgentCore + Google Workspace, DevRev MCP, etc.) frequently carry
JSON Schemas backed by legacy ``definitions`` (draft-04) or OpenAPI
``components.schemas`` instead of ``$defs`` (JSON Schema 2020-12).
Anthropic and Fireworks only resolve ``$defs``. Their tool-schema
filters silently drop the unrecognised def blocks while keeping the
``$ref`` pointers, so the upstream rejects the request:
- Anthropic: ``tools.0.input_schema: Invalid tool schema, $ref is
not supported``
- Fireworks: ``Error resolving schema reference '#/definitions/...'``
(PointerToNowhere)
Add ``unpack_legacy_defs(schema, *, copy=False)`` next to the existing
``unpack_defs`` -- a single helper that pops draft-04 ``definitions``
and OpenAPI ``components.schemas`` and feeds them through
``unpack_defs`` in place. ``$defs`` is left untouched (resolved
natively). ``copy=True`` deep-copies first when there is actually work
to do, used by Anthropic so the caller's tool dict is preserved.
Anthropic ``_map_tool_helper`` calls ``unpack_legacy_defs(_, copy=True)``;
Fireworks ``_transform_tools`` calls ``unpack_legacy_defs(params)``
in place.
Refs: https://github.com/BerriAI/litellm/issues/26692
Co-authored-by: Cursor <cursoragent@cursor.com>
|
||
|
Mateo Wang
|
65b6e04da6
|
fix: stop use_chat_completions_api flag from leaking into provider request body (#29447)
* fix: stop use_chat_completions_api flag from leaking into provider request body
use_chat_completions_api is a LiteLLM control flag that forces the
/responses -> /chat/completions bridge. It was missing from
all_litellm_params, so get_non_default_completion_params treated it as a
model-specific param and forwarded it to the upstream provider. A
model-level "use_chat_completions_api: true" in the proxy config therefore
reached the chat-completions path and was rejected by strict providers
(OpenAI/Anthropic) with HTTP 400 for an unknown body field.
Register it as a known internal param so it is stripped on every path
(completion, the responses bridge that calls litellm.completion, and
filter_out_litellm_params).
Adds a regression test driving litellm.completion() with a mocked OpenAI
client that asserts the flag never reaches the request body.
* test: clarify extra_body assertion in use_chat_completions_api leak test
Replace the misleading 'not in ... or {}' precedence idiom with an explicit
parenthesized guard that also handles extra_body being None.
|
||
|
Yassin Kortam
|
8190ff4d86
|
feat(otel): allowlist team_metadata sub-keys promoted to baggage (#29442) | ||
|
Yassin Kortam
|
fe108580d7
|
fix(datadog): split oversized batches on 413 instead of re-queueing forever (#29444) | ||
|
Mateo Wang
|
76bf280d0a
|
test(responses): bump deprecated gemini-3-pro-preview to gemini-3.1-pro-preview (#29433)
Google sunset gemini-3-pro-preview on the Gemini API, so the AI Studio
responses-API thought-signature tests started failing with a 404
("This model models/gemini-3-pro-preview is no longer available").
Point both tests at the current gemini-3.1-pro-preview model, which
litellm already has registered and which supports the function calling,
reasoning, and native streaming these tests exercise.
|
||
|
michelligabriele
|
80cf50dedb
|
fix(v3 limiter): cap no-max_tokens TPM floor at smallest configured limit (#28805) | ||
|
michelligabriele
|
117136cccc
|
fix(openai-moderation): wire streaming flags through to unified dispatcher (#27324) | ||
|
yuneng-jiang
|
7d1bd9d9f4
|
fix(reset_budget): write only {spend, budget_reset_at} and stop pre-zeroing counter (#29358)
* fix(reset_budget): write only {spend, budget_reset_at} and stop pre-zeroing counter
ResetBudgetJob's batched update_data path shipped the full key/user/team
model on each reset. Prisma rejects object_permission_id and budget_limits
on the update input type, so any row carrying those fields detonated the
entire batch -- spend never reset, budget_reset_at never advanced. After
v1.84.0 started populating object_permission_id on UI-created keys, this
fires routinely.
_reset_budget_common also zeroed the cross-pod spend counter before the
DB write, so failed resets left enforcement reading 0 from the counter
while the DB still held the over-budget spend, admitting requests past
the cap until the counter naturally re-saturated from new reservations.
Switch the write to per-row narrow updates ({spend, budget_reset_at})
via db.batch_, and move the counter invalidation out of
_reset_budget_common so it only fires after the DB write commits. On
DB-write failure the counter is left untouched, enforcement continues
to block, and the next scheduler tick can retry without leaving a
bypass window.
Fixes #27730.
* fix(reset_budget): address Greptile review on #29358
- Strengthen the bypass-half regression test: replace the for-loop over
call_args_list (vacuously true when empty) with assert_not_called(),
so the test would actually flag a re-introduction of counter-zeroing
via any code path.
- Add the same explanatory docstring on _write_user_reset_updates and
_write_team_reset_updates that _write_key_reset_updates already has,
so all three helpers point future maintainers at #27730.
* test(reset_budget): update test_proxy_budget_reset for new batch-write path
Same shape as the previous test_reset_budget_job.py update: keys/users/teams
now write through prisma.db.batch_().<table>.update, not update_data, so the
tests need a batcher mock and updated assertions. Adds:
- _wire_batcher_for_test helper that returns a list which accumulates per-row
batch updates captured from prisma_client.db.batch_().
- _attrify helper that wraps dict fixtures so getattr(item, "token") works
alongside the dict item-access the fake_reset_* mocks rely on. The new
narrow-write helpers use getattr to pull out the row's id, and would
silently skip plain dicts otherwise.
- Updates 3 partial_failure tests to assert against the batch-call list
(rows by id, payload contains only {spend, budget_reset_at}) instead of
update_data.assert_awaited_once + data_list inspection.
- Updates test_reset_budget_continues_other_categories_on_failure: only
budget + enduser still flow through update_data; key/user/team go through
the batch path now.
- Wires the batcher mock into 3 service_logger_*_success tests so commit()
is actually awaitable and the success hook fires.
These tests were silently passing locally only because the editable install
in .venv pointed at the main repo, not the worktree — running pytest with
PYTHONPATH overridden to the worktree (matching CI) reproduces the failures.
|
||
|
ryan-crabbe-berri
|
90b5104475
|
feat(mcp/auth): additive key access-group grants + opt-in member assignment (#29313)
* fix(mcp): make key.access_group_ids grants additive over team ceiling A key whose unified access_group_ids grant a private MCP server was having that grant intersected against its team's MCP ceiling, so a key in a team scoped to other servers (or with no own scope) lost the granted server entirely. Resolve access_group_ids once as ungated additive grants and union them on top of the key/team ceiling instead of folding them into the key scope that gets intersected. * test(mcp): align key access-group tests with additive-grant model The previous commit moved key.access_group_ids resolution out of the intersected key ceiling (_get_allowed_mcp_servers_for_key) and into the ungated additive grant path (_get_key_access_group_mcp_server_extras), unioned on top of the team ceiling. Five tests from #28890/#29195 still asserted the old gated / in-key-scope contract and failed: - _get_allowed_mcp_servers_for_key now returns the object_permission ceiling only and never resolves access_group_ids; two tests now assert the group resolver is not called from that path (with and without an object_permission present). - The extras path is ungated, so a group whose assigned_team_ids / assigned_key_ids exclude the caller still contributes its servers. - The end-to-end test asserts the grant surfaces via the extras path rather than the base key path. - Dropped test_key_access_group_ids_empty_returns_no_extras; the empty case is already covered by the extras family's no-groups test. * feat(auth): gate member access-group assignment on keys behind opt-in Non-admin team members could attach access_group_ids to keys they create or update, letting them self-grant resources (MCP servers/models) the team admin never intended. Add an opt-in KEY_ACCESS_GROUP_ASSIGNMENT team-member permission (default-deny) enforced at /key/generate and /key/update; proxy and team admins bypass. Surfaces automatically as a checkbox in the team Member Permissions UI. * fix(auth): gate access-group assignment on /key/regenerate too RegenerateKeyRequest inherits access_group_ids and prepare_key_update_data persists it, so a non-admin key owner could self-grant access groups by regenerating. Apply the same opt-in member gate using the existing key's team. * test(auth): cover member access-group gate and additive MCP grants Add unit tests for enforce_member_can_assign_access_groups (deny without opt-in, allow with opt-in, and proxy-admin / team-admin / non-team-key bypasses) and for _get_key_access_group_mcp_server_extras (no-auth and no-resolved-servers return empty, resolved ids are expanded, errors degrade to no grants). |
||
|
Shivam Rawat
|
dc4f5b12ef
|
fix(proxy): enforce allowed_passthrough_routes for auth=true pass-thr… (#29256)
* fix(proxy): enforce allowed_passthrough_routes for auth=true pass-through
Pass-through endpoints with auth=true were injected into openai_routes,
so teams with openai_routes access bypassed per-team allowed_passthrough_routes.
Gate auth-enforced pass-through at JWT, virtual-key, and non-admin route checks.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(proxy): clarify JWT passthrough denial
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(proxy): make pass-through auth checks method-aware
Prevent allowlist bypass when the same path is registered with different auth settings per HTTP method.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Fix passthrough route auth checks
* fix(proxy): reject unregistered pass-through HTTP methods
Enforce method-aware JWT checks and return 405 when stale FastAPI routes accept requests outside the current pass-through registry.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(proxy): remove duplicate request_method in JWT team lookup
Fixes SyntaxError on proxy startup caused by passing request_method twice to find_team_with_model_access.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Fix passthrough route auth enforcement
* fix(proxy): raise passthrough-specific 403 directly in virtual-key path
* fix(proxy): load team for RBAC role-claim JWT passthrough gating
* Revert "chore(tests): migrate Bedrock CI to AWS account 941277531214 (#28728)" (#29326)
This reverts the Bedrock CI account migration (#28728). The original account
(888602223428) was put under an AWS security restriction after a leaked key
and has since been reactivated, while the replacement account (941277531214)
lacks access to several models the suites exercise (legacy Bedrock Claude 3
models, Cohere, Nova Canvas image gen, Bedrock batch inference, and flagship
Opus). Pointing CI back at the reactivated account restores that coverage.
This is the exact inverse of #28728: all hardcoded 941277531214 references go
back to 888602223428 (provisioned/imported-model ARNs, AgentCore runtime ARNs
and their suffixes, batch execution role ARN, and the example proxy config),
the S3 buckets revert to litellm-proxy and load-testing-oct, the guardrail IDs
revert to wf0hkdb5x07f and ff6ujrregl1q, the SageMaker endpoint and Knowledge
Base revert to their original ids, and the live-call tests go back to the
legacy model strings. The grid_spec fail_reason workaround for the unentitled
Opus cells is dropped while keeping the unrelated bedrock_effort_ceiling field
added after the migration.
The CircleCI AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY env vars still point at
941277531214 and must be set to the reactivated account's fresh credentials
separately via the CircleCI API; AWS_REGION_NAME stays us-west-2.
(cherry picked from commit
|
||
|
Sameer Kankute
|
ba2699740c
|
feat(pass_through): extend passthrough_managed_object_ids to Azure (#29160)
* feat(pass_through): extend passthrough_managed_object_ids to Azure
Adds managed ID minting/resolution for Azure passthrough endpoints
(/azure/...) alongside the existing OpenAI passthrough support.
Key changes:
- pass_through_endpoints.py: detect azure/azure_ai custom_llm_provider
(string or enum) to set _is_managed_id_provider and _managed_id_provider;
both INPUT and OUTPUT rewrite blocks now fire for Azure.
- llm_passthrough_endpoints.py: forward custom_llm_provider into
create_pass_through_route so it reaches pass_through_request (was None).
- managed_id_rewriter.py: extend _PASSTHROUGH_PREFIX_RE and _canonical_path
to strip /azure/openai prefix and add /v1/ for Azure paths that omit it;
add ("azure", method, path) entries to BUILTIN_OUTPUT_ID_FIELD_MAP for
files and batches endpoints.
- managed_id_codec.py / types/utils.py: supporting codec and enum constant.
- proxy_server.py: register llm_passthrough_router before batches_router to
prevent route collision for /openai_passthrough/* paths.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(pass_through): remove unused imports for ruff F401
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(pass_through): satisfy mypy for optional parsed_body
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(pass_through): compute query params string after managed-ID rewrite
Move requested_query_params_str computation to after the managed-ID
input rewrite block so logging_url reflects the rewritten raw-provider
query params actually sent upstream, instead of the original
managed IDs.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* Add support for managed ids for passthrough responses api
* Add support for list batches and list files
* style: run Black on passthrough managed ID files
Fix CI formatting for managed_id_rewriter.py and pass_through_endpoints.py.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(passthrough): parse json file_object and implement before-cursor pagination
- Parse row.file_object via json.loads when Prisma returns it as a string;
mirrors openai_files_endpoints/common_utils.py so list responses keep all
stored detail fields (status, timestamps, etc.).
- Implement the previously-parsed-but-unused 'before' cursor for list
pagination by flipping fetch order to ascending with a 'gt' bound on
created_at, then reversing rows so the response stays newest-first.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* Remove logger
* refactor: split list_passthrough_ids_from_db to fix PLR0915
Extract pagination, fetch, and serialization helpers so the main list
function stays under the statement limit without changing behavior.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix: scope passthrough managed ID dedup and list by provider
Validate embedded provider before reusing deduped file/object rows so
OpenAI and Azure cannot share the same managed ID for an identical raw ID.
Filter list responses to rows whose managed IDs decode to the current
provider, with over-fetch scanning when needed.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(managed_id_rewriter): cap pagination trim at effective limit
When raw_limit > 100, fetch_limit is capped at 101 (one extra row to
detect has_more), but trimming with rows[:raw_limit] failed to drop
the sentinel row. Use the capped effective limit instead.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix: cross-provider object collision and fail-closed list error handling
Greptile P1: move provider check before access check in _mint_or_reuse_object
so a cross-provider raw ID collision (OpenAI and Azure share the same batch_
ID) falls through to mint a new provider-scoped row instead of raising 404.
Veria-ai medium: _fetch_provider_scoped_list_rows now always returns
(page, has_more) — DB errors break out of the scan loop and return matched
rows so far. list_passthrough_ids_from_db never returns None for a recognised
list route, so the caller can never fall through to the upstream provider.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix: namespace passthrough model_object_id by provider to prevent unique violation
Store model_object_id as 'passthrough:{provider}:{raw_id}' instead of the
bare raw ID so OpenAI and Azure can each own a row for the same raw batch ID
without hitting the @unique constraint. Dedup lookup uses the same namespaced
key so it is implicitly provider-scoped and the _managed_id_matches_provider
check is no longer needed on the object path.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix: gate list interception on managed_files hook like input/output rewrites
Without the hook no managed IDs are minted so the DB is empty. Intercepting
GET /v1/files without the hook returned an empty list and hid the caller's
real upstream files/batches. Matches the guard used by the input and output
rewrite blocks.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix: set has_more=True when scan cap is hit with a full final DB batch
When max_scans (20) is exhausted and the last DB page was full-sized,
there are almost certainly more rows beyond the scan window. Track
last_batch_full across iterations so the scan_cap_hit condition sets
has_more=True in that case, preventing silent pagination truncation in
high-mixed-provider pools.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(managed_id_rewriter): scope pagination cursor lookup to caller-owned rows
Prevent a cross-tenant timing oracle by constraining the after/before
cursor row lookup to the caller's owner_filter, and cover the real Azure
responses path form (no /v1/) in tests.
* fix(managed_id_rewriter): align passthrough list metadata with direct GET
Persist upstream file metadata when minting a managed file ID and rewrite
nested batch file IDs before snapshotting the object, so DB-served file/batch
list responses return the same fields and managed IDs as a direct endpoint
GET.
* fix(managed_id_rewriter): degrade to raw id on cross-owner object collision
The OUTPUT (mint) path of _mint_or_reuse_object raised HTTPException(404)
when a dedup hit on the namespaced model_object_id belonged to a different
owner, converting a successful upstream batch/response creation into a 404
for the caller. Two upstream accounts under one provider name can issue the
same raw id, so this is reachable in multi-tenant deployments.
Return the caller's raw id unmanaged instead: the upstream create already
succeeded, a new managed row can't be minted (model_object_id is @unique),
and reusing the other owner's managed id would later fail the access check.
* fix(managed_id_rewriter): scope list cursor by provider and cap body-rewrite recursion depth
* perf(managed_id_rewriter): push batch list provider scope to DB and anchor canonical-path prefix
Object (batch) list rows store model_object_id as passthrough:{provider}:{raw},
so the provider filter is now applied at the indexed DB column, collapsing the
application-layer multi-scan to a single query for that table. File rows keep
the decode-based scan since they have no provider column.
Anchor the canonical-path prefix regex at a path boundary so routes such as
/openai_realtime/... are no longer mis-stripped.
* fix(managed_id_rewriter): refresh stored batch snapshot on reuse
The dedup-reuse path in _mint_or_reuse_object returned the existing managed id
without updating the stored file_object, so DB-served list responses kept the
creation-time snapshot and showed null output_file_id/error_file_id even after
the batch completed. Refresh the snapshot when an owned row is reused so the
list reflects the batch's latest state.
* fix(managed_id_rewriter): deny cross-owner object access on retrieve/cancel/delete
Returning the raw id when can_access_resource fails only made sense for create
responses, where the caller's own upstream create succeeded under a raw id that a
different owner already holds. On retrieve/cancel/delete the caller reaches that
branch only by supplying another tenant's raw id (which bypasses the managed-id
input gate), so echoing the upstream object back leaked it cross-tenant. Restrict
the raw fallback to create routes and return 404 otherwise.
* fix(managed_id_rewriter): deny cross-owner file access on retrieve/delete
_mint_or_reuse_file scoped the raw file dedup lookup to the current caller, so a
raw file-... id belonging to another tenant was never found and the OUTPUT path
minted a fresh managed id for that same upstream file under the caller. A raw id
only reaches this path by skipping the managed-id input gate (raw provider ids
are opt-out), so a different-owner row means the caller is touching someone
else's file. Look up flat_model_file_ids globally and run can_access_resource;
deny with 404 on retrieve/delete and leave the raw id unmanaged on create, which
mirrors the cross-owner handling already in _mint_or_reuse_object.
* fix(managed_id_rewriter): deterministic provider-scoped file dedup
Replace the unscoped find_first in _mint_or_reuse_file with a find_many
ordered by created_at and an application-layer provider filter. The file
table has no provider column, so a raw file id shared across OpenAI and
Azure could map to one row per provider; find_first then picked a row
non-deterministically and, on a provider mismatch, minted a fresh managed
row on every call, accumulating duplicates. Selecting the oldest matching
same-provider row the caller can access keeps reuse stable and prevents
duplicate rows while preserving the cross-tenant deny/leave-raw behaviour.
* refactor(pass_through): scope passthrough managed IDs on the explicit provider
Move the openai/azure detection out of pass_through_request into
resolve_passthrough_managed_id_provider in llms/base_llm/managed_resources,
and key managed-ID rewriting on the forwarded custom_llm_provider rather than
the upstream URL's EndpointType. The helper documents why azure and azure_ai
collapse to one "azure" scope (they share the same Azure OpenAI files/batches
surface, so an ID minted on one must resolve on the other) and returns None for
any other provider so a third-party OpenAI-compatible endpoint never triggers
managed-ID minting.
Add TestManagedIdProviderScope covering the azure_ai -> azure collapse and the
non-openai/azure exclusion.
* test(log_db_metrics): assert sanitized event_metadata contract
test_log_db_metrics_success still asserted the legacy event_metadata
shape (function_name/function_kwargs/function_args), which #28909
intentionally removed so that live Prisma clients, OTel spans, and
secrets never land on a service-log span. The decorator now emits only
a sanitized payload: None when no table_name is present, and
{"table_name": ...} when it is. Update the test to verify both branches
of that contract.
* fix(managed_id_rewriter): page provider-scoped file list by offset
The file list scan advanced its cursor with a strict created_at boundary.
When several rows shared a created_at timestamp and a non-matching provider
row sat on the page boundary, the next query skipped the remaining rows at
that timestamp, dropping matching files from the response. Page by a stable
offset over a total order (created_at plus the unique id column) so tied
rows are never skipped or repeated.
* fix(managed_id_rewriter): push file-list provider scope to the DB
The file-list helper had no provider column to query, so it scanned the
table and filtered by decoding each managed ID in the application layer,
capped at 20 pages. For an admin with a large mixed-provider file pool
that cap could truncate a page.
Mint now writes a _passthrough_provider:{provider} marker into
flat_model_file_ids, giving the file table the same DB-queryable provider
scope object rows already get from the namespaced model_object_id. The
list helper pushes the scope into the query so a single round-trip serves
the page. The scan loop, the offset paging, and the cap are gone, so pages
can no longer truncate, leak the other provider, or skip rows that share a
created_at timestamp.
* fix(managed_id_rewriter): deny raw provider IDs that map to another tenant's managed resource
Clients only ever receive managed IDs on passthrough, so a raw file/batch/response ID for another tenant's managed object can only be recovered by decoding that tenant's managed ID. Raw IDs were forwarded upstream untouched (deliberate opt-out), which on a retrieve/cancel/delete executed upstream before the response-side ownership check ran, leaking a cross-tenant action.
Guard raw provider IDs on the input path: when a raw file-/batch_/resp_ ID resolves to a managed row the caller cannot access, return 404 before forwarding. Genuinely unmanaged raw IDs (no DB row) and IDs the caller owns are left untouched, preserving the opt-out.
* test(managed_id_rewriter): cover azure_ai and pre-versioned azure passthrough paths
* fix(managed_id_rewriter): fall back to raw id when persistence fails
A DB persistence failure after a successful upstream create left the
client holding a minted managed ID with no backing row, so every later
resolve returned 404 and the resource was permanently unreachable. Mint
the managed ID only when the row is stored; on persistence failure return
the raw provider id, matching the no-persistence-available fallback, so
the freshly-created resource stays reachable.
* fix(managed_id_rewriter): log only rewritten query param keys
* fix(managed_id_rewriter): use compound (created_at, id) list cursor boundary
A timestamp-only lt/gt cursor boundary skips list rows that share the
cursor row's created_at across a page boundary, silently dropping them.
Compare the unique id (the secondary sort key) alongside created_at so
the page walk stays complete when timestamps tie.
* fix(managed_id_rewriter): converge concurrent object creates on one managed id
* fix(managed_id_rewriter): bound raw-id guard DB lookups per request
The INPUT guard fired one DB lookup for every file-/batch_/resp_ prefixed
string in the path, query, and body. The file-id guard is an unindexed
array-containment scan over LiteLLM_ManagedFileTable, so an authenticated
caller could amplify a single passthrough request into thousands of
full-table scans by packing a body with id-shaped strings.
De-dupe raw ids within a request and cap the distinct guard lookups,
failing closed with 400 instead of skipping the guard. Legitimate callers
hold managed ids (resolved via an indexed unified_*_id lookup, not the
guard), so the cap only trips under abuse.
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
|
||
|
Shivam Rawat
|
7ca796beb1
|
fix(proxy): restrict vector store index create/delete to proxy admins (#29202)
* fix(proxy): restrict vector store index create/delete to proxy admins Prevent non-admin API keys from registering indexes via POST /v1/indexes or deleting Azure AI Search indexes through managed pass-through routes. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(proxy): tighten vector store index lifecycle checks Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
Shivam Rawat
|
4c3efe9c7c
|
fix(guardrails): return HTTP 400 for litellm content filter blocks (#28418)
* fix(guardrails): return HTTP 400 for litellm content filter blocks Align litellm_content_filter hard rejects with the standard guardrail block status code so clients receive 400 instead of 403. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(guardrails): return HTTP 400 for custom code guardrail blocks Pre-call custom code guardrail blocks now raise HTTPException(400) instead of using the passthrough ModifyResponseException path that returned a synthetic 200 response. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(guardrails): preserve custom code passthrough blocks Keep standalone custom code guardrail blocks on the passthrough contract while covering policy pipeline block handling for passthrough-style guardrail interventions. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
Mateo Wang
|
152b1177e5
|
test(reasoning-effort-grid): cover Claude Opus 4.8 across provider routes (#29327)
* test(logging): align DB metrics event_metadata assertions with safe redaction PR #28909 hardened log_db_metrics to emit a minimal, non-sensitive event_metadata (only table_name when present, otherwise None) instead of dumping function_name, function_kwargs, and function_args onto the span. The test in test_log_db_redis_services was not updated and still asserted "function_name" in event_metadata, which raised TypeError (argument of type 'NoneType' is not iterable) and turned the logging_testing CI job red on litellm_internal_staging. Update test_log_db_metrics_success to assert event_metadata is None when no table_name is passed, and add test_log_db_metrics_event_metadata_is_safe as a regression guard verifying that only the table name surfaces and that sensitive kwargs (tokens, prisma client) are never dumped. * test(bedrock): self-heal opus-4-7 grid cells when unentitled on CI The bedrock-claude-opus-4-7 converse cells are unentitled on the Bedrock CI account, so they were marked xfail. xfail keeps reporting them as expected failures even after access is granted, so the wire translation never gets verified again. Now the cell makes the call and skips only when Bedrock replies "is not available for this account"; the moment the model is entitled the same cells run their full assertions with no edit. A focused unit test pins the tolerance predicate so any other failure still surfaces loudly and the available path still runs the assertions. * test(reasoning-effort-grid): add claude-opus-4-8 across provider routes Adds claude-opus-4-8 to the anthropic, azure, vertex and bedrock-converse routes (275 cells total) so the reasoning-effort wire translation is covered for the new model. The bedrock opus-4-8 and opus-4-7 cells reuse the self-heal path: they run the call and skip only on Bedrock's "is not available for this account" reply, then assert in full once the model is entitled. The azure and vertex opus-4-8 cells stay xfail until a Foundry deployment exists and Vertex availability is confirmed. The shared xhigh+max capability set is renamed to _CAPS_XHIGH_MAX now that more than one model uses it. |
||
|
yuneng-jiang
|
94a043efb2
|
refactor(proxy/auth): normalize Bearer prefix in safe-hash helper (#29343)
* refactor(proxy/auth): normalize Bearer prefix in safe-hash helper
UserAPIKeyAuth._safe_hash_litellm_api_key now strips a leading
"Bearer "/"bearer " prefix before its existing sk-/JWT classification, so
the helper produces the same hashed output regardless of whether the
caller stripped the Authorization header prefix or passed the header
value through unchanged.
* refactor(proxy/auth): make Bearer-prefix strip case-insensitive
Per RFC 7235 the HTTP authorization scheme token is case-insensitive.
Replace the two-prefix loop with a single case-insensitive check so the
helper normalizes "Bearer ", "bearer ", "BEARER ", and any mixed-case
variant before classifying the remainder as sk- or JWT. The contract
test gains coverage of "BEARER " and "BeArEr ".
* test(mcp): align auth-handler test expectations with safe-hash helper
The two MCP auth tests asserted that UserAPIKeyAuth(api_key="Bearer ...")
retained the raw header bytes on the api_key field. _safe_hash_litellm_api_key
now normalizes that input — stripping the Bearer prefix and hashing the
resulting sk- key — so the expectations move to the normalized form:
the bare token in the parametrize case, and hash_token("sk-...") in the
backward-compat assertion. This matches what the real auth flow produces
(the builder strips Bearer and the DB stores the hashed token), so the
mocks now line up with production rather than with the un-normalized
validator output.
|
||
|
Mateo Wang
|
bfbb5d2375
|
fix(ci): make litellm_internal_staging green (logging test + Bedrock Opus 4.7 self-heal) (#29344)
* test(logging): align DB metrics event_metadata assertions with safe redaction PR #28909 hardened log_db_metrics to emit a minimal, non-sensitive event_metadata (only table_name when present, otherwise None) instead of dumping function_name, function_kwargs, and function_args onto the span. The test in test_log_db_redis_services was not updated and still asserted "function_name" in event_metadata, which raised TypeError (argument of type 'NoneType' is not iterable) and turned the logging_testing CI job red on litellm_internal_staging. Update test_log_db_metrics_success to assert event_metadata is None when no table_name is passed, and add test_log_db_metrics_event_metadata_is_safe as a regression guard verifying that only the table name surfaces and that sensitive kwargs (tokens, prisma client) are never dumped. * test(bedrock): self-heal opus-4-7 grid cells when unentitled on CI The bedrock-claude-opus-4-7 converse cells are unentitled on the Bedrock CI account, so they were marked xfail. xfail keeps reporting them as expected failures even after access is granted, so the wire translation never gets verified again. Now the cell makes the call and skips only when Bedrock replies "is not available for this account"; the moment the model is entitled the same cells run their full assertions with no edit. A focused unit test pins the tolerance predicate so any other failure still surfaces loudly and the available path still runs the assertions. |
||
|
Shivam Rawat
|
2d4c13c00f
|
fix(mcp): preserve omitted fields on PUT /v1/mcp/server partial updates (#29253)
* fix(mcp): preserve omitted fields on PUT /v1/mcp/server partial updates Use model_dump(exclude_unset=True) for updates so schema defaults (transport=sse, allow_all_keys=false, etc.) are not written when callers omit them. Serialize JSON fields from the filtered dict and only force is_byok on create. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(mcp): allow explicit alias=None on MCP server partial updates Snapshot caller-provided fields before normalization so omitted alias is not written while an intentional alias=None still clears the stored value. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
Shivam Rawat
|
ace3c65ab3
|
fix(mcp): preserve source_url in GET /v1/mcp/server list responses (#29249)
* fix(mcp): preserve source_url in GET /v1/mcp/server list responses
The list endpoint builds responses from the in-memory registry, but
source_url was dropped during the DB-to-registry roundtrip even though
GET /v1/mcp/server/{id} returned it correctly from the database.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(tests/mcp): set source_url on MagicMock table records
MagicMock auto-creates source_url as a mock object, which fails MCPServer
Pydantic validation after source_url was wired through build_mcp_server_from_table.
Co-authored-by: Cursor <cursoragent@cursor.com>
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
|
||
|
Mateo Wang
|
f11c12d157
|
Revert "chore(tests): migrate Bedrock CI to AWS account 941277531214 (#28728)" (#29326)
This reverts the Bedrock CI account migration (#28728). The original account (888602223428) was put under an AWS security restriction after a leaked key and has since been reactivated, while the replacement account (941277531214) lacks access to several models the suites exercise (legacy Bedrock Claude 3 models, Cohere, Nova Canvas image gen, Bedrock batch inference, and flagship Opus). Pointing CI back at the reactivated account restores that coverage. This is the exact inverse of #28728: all hardcoded 941277531214 references go back to 888602223428 (provisioned/imported-model ARNs, AgentCore runtime ARNs and their suffixes, batch execution role ARN, and the example proxy config), the S3 buckets revert to litellm-proxy and load-testing-oct, the guardrail IDs revert to wf0hkdb5x07f and ff6ujrregl1q, the SageMaker endpoint and Knowledge Base revert to their original ids, and the live-call tests go back to the legacy model strings. The grid_spec fail_reason workaround for the unentitled Opus cells is dropped while keeping the unrelated bedrock_effort_ceiling field added after the migration. The CircleCI AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY env vars still point at 941277531214 and must be set to the reactivated account's fresh credentials separately via the CircleCI API; AWS_REGION_NAME stays us-west-2. |
||
|
Shivam Rawat
|
967fed1fa1
|
feat(enterprise): add RESEND_FROM_EMAIL for self-hosted Resend sends (#28830)
Allow self-hosted installs to override the default LiteLLM sender address via RESEND_FROM_EMAIL, matching SendGrid's SENDGRID_SENDER_EMAIL pattern. Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
Sameer Kankute
|
4cc3dd7aad
|
feat(context_management): compact_20260112 polyfill for non-Anthropic providers (#28868)
* feat(anthropic/messages): in-gateway context_management polyfill for non-Anthropic providers
- Add `context_management/` module with `clear_tool_uses_20250919` editor
dispatched before chat-completions translation on `/v1/messages`
- Hard-protect most-recently completed tool_result from being cleared
- Attach `context_management.applied_edits` to both non-streaming and
streaming (final `message_delta`) responses
- Bedrock Converse: forward `context_management`; filter to
`compact_20260112`-only edits with `compact-2026-01-12` beta header
- token_counter: guard Anthropic-format tools (no `function` key) to
prevent AttributeError during polyfill token counting
- Streaming: handle empty-choices usage-only trailing chunks
- Skip polyfill when `litellm.drop_params = True`
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(bedrock): pop None context_management before sending to Bedrock Converse
If context_management is forwarded as None (e.g. when mapping returns
None for an invalid format), _filter_context_management_for_bedrock_converse
previously returned early without removing the key, leaving
"context_management": null in the request and causing a validation
error. Pop the key when the value is not a dict.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(bedrock/converse): pop None context_management; extract helpers to fix PLR0915
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(anthropic/messages): check per-request drop_params alongside global
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(anthropic/messages): preserve drop_params for downstream and respect explicit False
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix: lazy debug logging in clear_tool_uses; remove unused context_management constants
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(anthropic/messages): guard context_management polyfill with try/except
Wrap apply_context_management() in a try/except so any failure (e.g.
litellm.token_counter raising on an unknown tokenizer or unexpected
message format) is logged but does not crash the underlying LLM
request. The polyfill is a best-effort additive feature; on failure we
forward the original messages without applied edits.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(token_counter): guard None input_schema in Anthropic tool fallback
Use `or {}` instead of `.get(..., {})` so explicit null parameters do not
raise AttributeError when formatting function definitions for token counting.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix: minimize context_management polyfill threading
- Use None (not empty list) for polyfill_applied_edits when context
management isn't requested, so semantics of 'feature not requested'
vs 'feature requested but no edits applied' are distinct.
- In the streaming iterator, only pass applied_edits to the per-chunk
translator on the final (finish_reason) chunk; intermediate chunks
ignore it anyway, and this makes intent explicit on both sync and
async paths.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(context_management): align tool_use counts and normalize list spec
- _count_tool_uses now requires a string id, matching _collect_tool_use_ids_in_order so the tool_uses trigger can't fire on blocks that aren't clearable.
- apply_context_management dispatcher now accepts the OpenAI list form and normalizes it via AnthropicConfig.map_openai_context_management_to_anthropic, so the polyfill path no longer silently no-ops on list input.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* feat(context_management): add compact_20260112 polyfill for non-Anthropic providers
Implements an in-gateway compaction polyfill that summarizes long conversations
using a configurable model when `compact_20260112` is requested for non-Anthropic
targets (e.g. OpenAI, Gemini), matching Anthropic's context management beta
behaviour for those providers.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(compact): skip tool_result-only user turns; bedrock: elif for context_management
- compact_20260112 Phase D: when keeping the last user turn after a full
summary, skip role=user turns whose content is exclusively tool_result
blocks. Such turns translate to OpenAI tool-role messages with no
preceding assistant tool_calls (those got summarized away), which
non-Anthropic providers reject. Fall back to a synthetic continuation
prompt if no eligible user question exists, so the downstream call
always has a non-empty user message.
- bedrock converse: chain the context_management param as elif so it
follows the same if/elif pattern as the surrounding thinking/
reasoning_effort checks.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(anthropic): post-compaction question selection, system type, sync stream merge
- compact.py: select last user question from effective_messages (post-compaction slice) instead of raw messages, so prior summarized turns aren't reintroduced
- handler.py: widen _prepare_completion_kwargs system parameter type to Union[str, List[Dict]] matching PolyfillResult.system
- streaming_iterator.py: mirror async hold-and-merge logic in sync __next__ so context_management is attached to the final merged message_delta when stop_reason and usage arrive in separate chunks
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(anthropic/messages): apply context_management on sync path; clear held stop_reason chunk in async iterator
- Sync `anthropic_messages_handler` was silently dropping the
`context_management` kwarg via `ANTHROPIC_ONLY_REQUEST_KEYS` after the
polyfill was moved into the async handler. Bridge to the async
dispatcher with `run_async_function` so `litellm.messages.create()`
callers keep working (regressed e.g. `clear_tool_uses_20250919`).
- In the streaming iterator's `__anext__` `StopIteration` handler, clear
`self.holding_stop_reason_chunk` after capturing it (matches `__next__`)
so a subsequent call doesn't re-emit the same chunk.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(bugfixes): bedrock None context_mgmt; stream per-instance queue; sync polyfill; trailing-chunk passthrough
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(anthropic): silently drop trailing chunks after usage; remove dead _polyfill_result key
- streaming_iterator: in sync __next__, after the usage chunk has been
merged and emitted, silently consume any trailing provider events
via 'continue' instead of forwarding them through the queue. Trailing
chunks would translate to content_block_delta or message_delta and
violate Anthropic SSE ordering after the final message_delta. The
async __anext__ already drops these via 'if not self.queued_usage_chunk:'
gating, so this aligns sync and async behavior.
- handler: drop unused '_polyfill_result' from ANTHROPIC_ONLY_REQUEST_KEYS.
PolyfillResult is passed as an explicit arg to the adapter methods, never
through extra_kwargs, so the entry was dead code.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* refactor(anthropic): extract usage-merge helper; guard empty slice-only compaction result
- Extract the duplicated hold-and-merge usage logic from the sync __next__ and
async __anext__ paths into a shared _merge_usage_into_held_stop_reason_chunk
helper so the subtle cache-token / context_management attachment lives in
exactly one place.
- In the compact_20260112 slice-only path, fall back to _select_last_user_question
when _strip_compaction_blocks produces an empty list (e.g. messages ending on
an assistant turn whose only content was the compaction block) so the
downstream API never receives an empty messages array.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* refactor(anthropic/context_management): streaming iterator compaction fixes and compact polyfill improvements
- Extract usage-merge helper; guard empty slice-only compaction result
- Silently drop trailing chunks after usage; remove dead _polyfill_result key
- Fix bedrock None context_mgmt; stream per-instance queue; sync polyfill; trailing-chunk passthrough
- Apply context_management on sync path; clear held stop_reason chunk in async iterator
- Fix post-compaction question selection, system type, sync stream merge
- Skip tool_result-only user turns; bedrock: elif for context_management
- Add streaming iterator compaction test suite
Co-authored-by: Cursor <cursoragent@cursor.com>
* revert(html): restore flat *.html naming in _experimental/out
Reverses the accidental rename from *.html → */index.html introduced in
15ea941fbe. All 35 files moved back to their original flat paths so the
directory structure matches litellm_internal_staging.
Co-authored-by: Cursor <cursoragent@cursor.com>
* revert(config): restore proxy_server_config.yaml to litellm_internal_staging
Co-authored-by: Cursor <cursoragent@cursor.com>
* Fix: skip client compaction pre-processing when compact_20260112 polyfill will run
The _prepare_context_managed_request helper unconditionally applied
apply_client_compaction_block_history before invoking the polyfill. When
the request also configured a compact_20260112 spec, that pre-processing
consumed the client-sent compaction block and collapsed the message history
to just the latest user question, starving the polyfill of conversation
context. The polyfill's own Phase A (_slice_around_compaction_block)
already handles client compaction blocks correctly and inspects the full
post-compaction tail for the token-threshold check, so the pre-processing
is both redundant and destructive in this case.
Now the pre-processing only runs when no compact_20260112 polyfill spec
will execute (no spec, drop_params on, or only non-compact edits like
clear_tool_uses_20250919).
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(anthropic): plug compaction-block leak + iteration-usage gaps in streaming adapter
- handler: when polyfill_will_run skipped client-history pre-processing
and the polyfill ultimately returned None (best-effort swallow on
unexpected error), apply the slice-only fallback before returning so
Anthropic-specific 'compaction' content blocks don't leak to non-
Anthropic backends that would reject them.
- streaming_iterator: precompute will_merge_into_held so we don't pass
applied_edits into the translator when the resulting processed_chunk
will be discarded by the held stop-reason merge path.
- streaming_iterator: augment processed_chunk with iterations usage in
the holding_chunk branch (sync and async) for parity with the other
emission branches; ensures usage.iterations is attached on the rare
message_delta-reaches-holding_chunk path.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(anthropic): correct streaming usage iteration + translate tools for token counting
- streaming_iterator: skip the trailing "message" iteration entry in the
final message_delta when the held stop_reason chunk carries placeholder
zero usage (no separate usage chunk arrived). Reporting zero tokens was
misleading and inconsistent with the non-streaming path which always
has real usage data.
- streaming_iterator: drop two redundant type checks inside branches
that are already guarded by an outer message_delta type check.
- compact._count_effective_tokens: translate Anthropic-shaped tools
(input_schema) to OpenAI shape before passing to litellm.token_counter
so threshold checks aren't skewed by tokenizer paths that expect the
OpenAI tool wrapper.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* Fix lint
* fix(anthropic): plug content drop, compaction SSE shape, and compaction leak
- Sync streaming __next__ no longer drops a buffered holding_chunk when
the usage-merge path has already fired. Restoring the prior unconditional
flush behavior preserves provider-emitted content (the SSE-ordering nit
of a trailing content delta is preferable to silent content loss).
- compaction content_block_start now carries the full block shape
({"type": "compaction", "content": ""}) to match the text-block
pattern and Anthropic's native streaming shape, so clients that key off
content_block_start see the field.
- apply_compact_20260112 now slices around / strips compaction blocks
before the opt-in gate check. Previously, when summary_model was not
configured the editor returned the raw messages, leaking Anthropic-only
compaction content blocks to non-Anthropic providers that reject them.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(anthropic): resolve mypy types in context management polyfill
Use AppliedEdit and CompactionBlock consistently in the dispatcher and streaming adapter.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(anthropic): flush held content chunk in async streaming path
Mirror the sync __next__ behavior: always flush a buffered
holding_chunk after the stream ends, even when usage was already
merged + emitted. Previously the async __anext__ kept the flush
inside the 'if not self.queued_usage_chunk:' guard, silently
dropping the last content delta on the proxy's primary path.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(anthropic adapter): correct sync streaming, surface polyfill failures, decouple sync path from proxy router
- translate_completion_output_params_streaming: add is_async flag so the
sync handler returns Iterator[bytes] instead of an unusable
AsyncIterator. Async callers keep the existing behavior via the
default is_async=True.
- _run_polyfill_if_enabled: when the polyfill crashes and the spec
requested non-compact edits (e.g. clear_tool_uses_20250919), raise an
AnthropicContextManagementError instead of silently returning None so
those edits are not dropped without an error surface. The
compaction-block-slicing safety net remains for compact-only specs.
- anthropic_messages_handler (sync): stop auto-attaching the proxy
llm_router. run_async_function bridges to a new thread's event loop;
reusing the proxy's loop-bound httpx clients there causes
'Event loop is closed' errors. The summary editor falls back to
litellm.acompletion when llm_router is None.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix: address bug detection findings in token counter and streaming iterator
- token_counter: guard against non-dict 'function' field in tool dicts
and skip tools missing a name to avoid emitting 'type None = ...' which
would produce inaccurate token counts.
- streaming_iterator: change sync __next__ generic-error path to raise
StopIteration (was StopAsyncIteration), so sync iteration cleanly stops.
- streaming_iterator: centralize context_management attachment so the
held-stop_reason direct-flush path defensively re-attaches applied_edits
to match the merge path's guarantee.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* Fix lint
* fix: correct COMPACT_MIN_TRIGGER_TOKENS to 50_000
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* Fix lint
* Fix lint
* Fix lint
* fix(compact): reduce to last user question when summary_model not configured but prior compaction block exists
Aligns the summary_model_not_configured path with the under-threshold and
client-compaction-block paths, which both reduce post-compaction messages
to just the latest user question so the downstream provider doesn't get
the summary on system prefix AND the full post-compaction history.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(compact): forward caller system prompt to summary model call
The default summarization instructions reference "the initial task above"
and "the raw history above", but the system prompt that holds that task
was not being forwarded to the summary model. The summary call now
prepends an OpenAI-shaped system message translated from the original
Anthropic-shaped system (str or content-block list) so the summarizer
has the agent role and initial task in scope.
* fix(compact_20260112): set default max_tokens and merge prompt when last turn is user
- Set COMPACT_SUMMARY_MAX_TOKENS default for the summary call so providers
like Anthropic (which require max_tokens) don't silently fail and degrade
to summary_call_failed.
- When the trailing translated message is already a user turn, merge the
summarization prompt into it instead of appending a second user turn.
Avoids consecutive role=user messages that strict providers reject.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(anthropic adapter): move current_content_block_start to __init__
Move the default TextBlock dict from a class-level attribute to __init__ so
concurrent stream instances don't share the same mutable dict. The class-level
default could be mutated in-place via tool_block['name'] = original_name in
_should_start_new_content_block, leaking state across streams. This mirrors
the existing fix already applied to chunk_queue.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(compact_20260112): surface error states + strip tool_result blocks in last user question
applied_edits_for_response() now includes compact_20260112 edits that
carry an error field (summary_model_not_configured, summary_call_failed,
summary_extraction_failed) so clients and operators can see why
compaction was requested but not applied.
_select_last_user_question() now strips tool_result blocks from mixed
[tool_result, text] turns rather than passing them through as-is. After
compaction the paired tool_use assistant turn no longer exists, so
forwarding tool_result blocks translates to orphaned role=tool messages
on non-Anthropic providers and produces a 400.
* fix(compact_20260112): carry prior compaction summary into Phase C summary call
When a request already contains a compaction block, Phase A slices
`effective_messages` to the turns since that block. Previously Phase C
passed the original `system` to the summary model, so multi-round
compaction silently dropped accumulated history each time the polyfill
fired. Pass `augmented_system` (original system + prior summary
prefix) so the summary model can produce a comprehensive summary that
incorporates both the prior round's context and the current slice.
`summarized_system` for the downstream call stays built from the
original `system` + new `summary_text`.
* refactor: delegate handler spec normalization to dispatcher
_normalize_spec_edits in adapters/handler.py duplicated the spec-shape
normalization already implemented by _normalize_spec in
context_management/dispatcher.py. The two could drift: a change in one
(e.g. supporting a new spec shape) without the other would cause the
handler's polyfill_will_run prediction to disagree with the
dispatcher's actual behavior, breaking the client-history pre-processing
skip.
Have the handler delegate to the dispatcher's _normalize_spec while
keeping handler-specific concerns (drop_params short-circuit, swallow
mapping exceptions) at the wrapper level.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(compact_20260112): surface warning-only applied edits in response
`applied_edits_for_response()` previously hid `compact_20260112` edits when
they had only warnings (no compaction block, no error). This dropped
diagnostically important warnings such as
`unsupported_trigger_type_X_using_input_tokens` and
`pause_after_compaction_ignored` whenever the conversation was under the
trigger threshold. Operators now see these warnings in the response.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix: address two low-severity context_management edge cases
- streaming_iterator: keep `sent_content_block_finish` in sync with the
compaction block's emitted start/delta/stop lifecycle and reset it when
the next text block's start is queued.
- bedrock _map_context_management_param: match dispatcher `_normalize_spec`
behavior — only run the OpenAI→Anthropic mapper on list inputs; pass
dict inputs through unchanged so already-Anthropic-format values aren't
silently dropped.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(compact_20260112): use beta-header constant; require type discriminator; skip sync bridge when idle
- bedrock: replace hardcoded "compact-2026-01-12" beta string with
ANTHROPIC_BETA_HEADER_VALUES.COMPACT_2026_01_12.value in both
Converse (_filter_context_management_for_bedrock_converse) and Invoke
(anthropic_claude3) compact-edit handlers.
- types: mark the "type" discriminator as Required[...] on the new
CompactionBlock and UsageIteration TypedDicts so the discriminator
is not silently optional under total=False.
- adapters/handler: short-circuit the sync /v1/messages adapter path
before spawning the run_async_function worker-thread event loop when
the request has no context_management spec and no client-sent
compaction block in the message history.
Test plan:
- uv run pytest tests/test_litellm/llms/anthropic/experimental_pass_through/ tests/test_litellm/llms/bedrock/test_converse_context_management.py -q
(370 + 10 = 380 passed)
- uv run pytest tests/test_litellm/llms/azure_ai/claude/test_azure_anthropic_transformation.py tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_transformation.py -k compact (3 passed)
* fix(compact_20260112): include system prompt tokens in threshold check
The threshold check in Phase B previously counted only message tokens and
the compaction-block content, omitting the system prompt entirely. When
the system carried a prior compaction summary (via _augment_system_with_summary)
or was otherwise large, the threshold could fire later than intended,
allowing the conversation to exceed the model's context window before
compaction activated.
_count_effective_tokens now also counts the (augmented) system prompt
text. The caller passes compaction_block=None when augmented_system
already includes the prior summary, to avoid double-counting.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* Fix SSE ordering and compaction state machine bugs in AnthropicStreamWrapper
- Suppress holding_chunk flush after final message_delta has been emitted
(queued_usage_chunk == True) so a trailing content_block_delta cannot
follow message_delta, which strict Anthropic SDK clients may reject.
When usage has not yet been merged, flush the holding_chunk *before*
the held stop_reason chunk so SSE ordering remains correct.
- Replace _queue_compaction_block_events with _next_compaction_event,
emitting the compaction start/delta/stop events one at a time. The
state machine flags (sent_content_block_finish) and content block
index now advance atomically with the terminal stop event actually
being returned to the caller, eliminating the transient inconsistent
state where flags say the block is finished while its stop event is
still buffered.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(compact_20260112): enforce parent key/team allowlist on summary model
The compact_20260112 polyfill summary subrequest used llm_router.acompletion
directly, bypassing the proxy auth checks that gate model access for the
parent key/team. A caller whose key/team was not authorized for the
configured context_management_summary_model could still cause the proxy to
invoke that model and return its output as a compaction block.
Pull the parent's UserAPIKeyAuth out of litellm_metadata in the handler,
thread it through the dispatcher into apply_compact_20260112, and gate the
summary call on _can_object_call_model for both key-level and team-level
allowlists. Failures land as applied_edits[0].error =
summary_model_access_denied without raising. SDK callers (no UserAPIKeyAuth)
remain unaffected.
* fix(compact_20260112): distinguish access-denied from transient errors; greedy summary regex
- _check_summary_model_access now catches ProxyException explicitly for access
denials and logs unexpected exceptions separately. Both still fail closed,
but operators can now tell a denied key/team apart from a router internal
raising during the check.
- _SUMMARY_TAG_RE switches from non-greedy to greedy so a stray </summary>
inside the model's summary content no longer silently truncates the
captured text.
* fix(compact_20260112): type object_type as Literal for mypy
* fix(compact_20260112): attribute summary subcall spend to parent key/team
The compact_20260112 polyfill summary subrequest propagated metadata via
the Anthropic-shape `metadata` parameter, which only carries `user_id`.
The proxy auth fields used for spend attribution (`user_api_key`,
`user_api_key_team_id`, `litellm_call_id`, ...) live in
`data["litellm_metadata"]`. As a result, summary subcalls landed on the
router with an empty propagated metadata and the resulting tokens were
not attributed to the caller's key/team budget.
Rename the polyfill chain's spend-propagation parameter to
`litellm_metadata` and pull it from `kwargs["litellm_metadata"]` in
both the async and sync handlers, so the post-call hooks see the parent
key/team and bill the summary tokens accordingly. Add an
`_extract_proxy_litellm_metadata` helper and refactor
`_extract_user_api_key_auth` to use it.
* chore(anthropic adapters): remove unused _extract_user_api_key_auth helper
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* chore(compact_20260112): non-greedy summary regex; use COMPACT_EDIT_TYPE in bedrock filter
- Make _SUMMARY_TAG_RE non-greedy so a response with multiple <summary>
blocks captures only the first complete block.
- Replace the hardcoded 'compact_20260112' literal in
_filter_context_management_for_bedrock_converse with the shared
COMPACT_EDIT_TYPE constant.
* fix: bug fixes from PR review
- streaming_iterator: don't set sent_content_block_finish during compaction
block lifecycle; that flag tracks the regular text/tool_use/thinking block
state machine, conflating the two leaks bad state to introspection paths.
- compact._call_summary_model: send propagated proxy auth/spend-attribution
fields as 'litellm_metadata' instead of 'metadata' so the router's post-call
hooks attribute summary tokens to the caller's key/team budget.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(anthropic-streaming): insert content_block_stop between held delta and final message_delta
When the stream exhausts with both `holding_chunk` (a content_block_delta)
and `holding_stop_reason_chunk` (a message_delta) buffered, the after-loop
cleanup previously emitted them back-to-back, producing the invalid
Anthropic SSE sequence `content_block_delta -> message_delta`. Insert a
`content_block_stop` between them in both the sync `__next__` and async
`__anext__` paths so the emitted ordering remains
`content_block_delta -> content_block_stop -> message_delta`.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(compact_20260112): propagate allowed_model_region to summary subrequest
The router enforces region restrictions by reading allowed_model_region
from top-level request kwargs (Router._common_checks_available_deployment),
but the compact_20260112 summary subrequest only forwarded litellm_metadata.
A region-restricted caller could trigger compaction and have their conversation
summarized by a deployment outside the permitted region.
Extract allowed_model_region from user_api_key_auth and pass it through
_call_summary_model as a top-level kwarg so the router applies the same
region constraints the parent request would.
* fix(anthropic adapter): emit content_block_stop before held message_delta in drain paths
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* feat(context_management): configurable summary max_tokens; surface ignored knobs
- compact_20260112: read summary max_tokens from general_settings
(context_management_summary_max_tokens) so operators can fit the
chosen summary model's output budget; falls back to the compiled
default for missing or invalid values.
- clear_tool_uses_20250919: log unsupported knobs at warning level
(was debug, which silently dropped misconfiguration) and surface
them as warnings on the AppliedEdit so clients see what was ignored.
* fix(compact_20260112): bound _call_summary_model with timeout
A slow or unresponsive summary model previously hung the parent
/v1/messages request with no escape hatch. Pass a 60s timeout on the
litellm.acompletion / llm_router.acompletion subrequest; on timeout the
existing summary_call_failed path forwards the request without
compaction rather than blocking indefinitely.
* fix(compact_20260112): preserve post-compaction tail on slice-only path
When a prior compaction block is present and the request is under threshold,
the polyfill was reducing downstream messages to just the latest user
question. The prior summary only covers turns before the compaction block,
so dropping the post-compaction tail silently lost recent context — a
multi-turn conversation that stayed below the threshold would arrive at the
model with no memory of any turn after the prior compaction.
Forward the already-stripped post-compaction tail unchanged on both the
under-threshold path and apply_client_compaction_block_history. Fall
back to _select_last_user_question only when the strip leaves nothing
for the downstream call to answer.
* fix(compact_20260112): enforce user/project/team-member model scopes on summary subrequest
The local gate previously only checked the parent key's and team's
allowed-model lists. A caller restricted by a personal user, project,
or per-team-member allowed_models scope could still trigger the
configured summary model and receive its <summary> output as a
compaction block, because llm_router.acompletion bypasses the proxy
common_checks path. Extend _check_summary_model_access to also load
the user_object, project_object, and team_membership and run the
matching allowlist check at each scope before invoking the summary
model.
* fix(compact_20260112): enforce summary model per-model budget and propagate budget metadata
* fix(compact_20260112): forward post-compaction tail when summary model unconfigured
* fix(anthropic endpoints): run failure hook on 500-level context management errors
* fix(compact_20260112): enforce summary model rate limit before summary call
* fix(compact_20260112): propagate end-user/project budget scope to summary call
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
|
||
|
Yassin Kortam
|
3be3c1dea1
|
feat(otel): add team_metadata, http.route, and model names to inference spans (#29319) | ||
|
yuneng-jiang
|
12d29a38a7
|
tests(proxy_server): surface current behavior in tests (#29309)
* test(proxy/proxy_server): pin forwarding routes (PR2) (#28887) * test(proxy): pin proxy_server.py forwarding-route behavior PR2 of the proxy_server.py behavior-pinning project: fills the 12 forwarding-route test files added by the harness PR with happy + error pins for all 52 LLM-facing routes (models, chat/completions, completions, embeddings, moderations, audio, assistants, threads, utils, model-info, model-metrics, queue). Every happy-path test asserts the full response dict via normalize() so the gate enforces real shape pinning rather than status codes. * test(proxy): drop task-plumbing comments from PR2 test files * test(proxy): tighten PR2 error-path status-code pins Apply the same review feedback Greptile gave on PR1 (#28856) and PR3 (#28850) to PR2's forwarding-route tests: - Replace permissive `>= 400` / `in (X, Y)` status assertions with the exact 500/405 the handler actually returns, so a regression that silently shifts the code now fails the pin. - Add a body-presence check alongside each tightened status assertion to satisfy _pin_check.py's no-status-only rule. --------- Co-authored-by: Claude <noreply@anthropic.com> * test(proxy): pin proxy_server.py non-route surface behavior (PR1) (#28856) * test(proxy): pin proxy_server.py non-route surface behavior (PR1) Fills the 7 PR1 placeholder files under tests/test_litellm/proxy/proxy_server/ with behavior pins for the non-route surface of proxy_server.py: lifecycle/init/shutdown, ProxyConfig class methods, DB-overlay config scrubbers, spend counters, background-health helpers, OpenAPI customization, exception handlers, and streaming-generator helpers. 233 tests cover 101 pin-list symbols (1+ happy + 1+ error each). New-tests-only coverage on litellm/proxy/proxy_server.py: 32.80% line / 20.91% branch (PR1 gate: 25% line / 18% branch). Full directory runs in ~22s with -n 4. Plan: https://www.notion.so/Plan-Pin-proxy_server-py-behavior-2026-05-25-36c43b8acdab81ee845fd5365128a2fc * test(proxy): address Greptile review comments on test_lifecycle.py - test_initialize_signature_is_async_with_expected_params: hard-code expected_param_count so a signature change actually trips the gate (previously both sides of the comparison were len(sig.parameters)). - test_check_request_disconnection_invalid_when_connected_times_out: patch asyncio.sleep so the test no longer spins for ~1.2 s of real wall-clock; timeout lowered to 0.05 s. --------- Co-authored-by: Claude <noreply@anthropic.com> * test(proxy/proxy_server): pin control-plane routes (PR3) (#28850) * test(proxy/proxy_server): pin misc routes (PR3, partial) Adds happy + error tests for the misc control-plane routes: GET /, /routes, /adaptive_router/state, /get_logo_url, /get_image, /get_favicon. Also gitignores .pin_list.txt (used by the pin gate). * test(proxy/proxy_server): pin login/SSO routes (PR3, partial) Adds happy + error tests for the 5 login/SSO control-plane routes: GET /fallback/login, POST /login, POST /v2/login, POST /v3/login, POST /v3/login/exchange. Mocks authenticate_user and create_ui_token_object at their imported location. * test(proxy/proxy_server): pin onboarding routes (PR3, partial) Adds happy + error tests for the 2 onboarding control-plane routes: GET /onboarding/get_token, POST /onboarding/claim_token. Wires a MagicMock async context manager for prisma_client.db.tx() and signs the onboarding JWT with the patched master_key. * test(proxy/proxy_server): pin model_cost_map reload routes (PR3, partial) Adds happy + error tests for the 5 model-cost-map control-plane routes: POST /reload/model_cost_map, POST|DELETE|GET /schedule/model_cost_map_reload(/status), GET /model/cost_map/source. Attaches litellm_config to mock_prisma per-test (the table is not in the default _PRISMA_TABLES fixture). * test(proxy/proxy_server): pin anthropic_beta_headers reload routes (PR3, partial) Adds happy + error tests for the 4 anthropic-beta-headers control-plane routes: POST /reload/anthropic_beta_headers, POST|DELETE|GET /schedule/anthropic_beta_headers_reload(/status). Stubs db.litellm_config (not in default _PRISMA_TABLES) and monkeypatches reload_beta_headers_config so no network calls fire. * test(proxy/proxy_server): pin invitation routes (PR3, partial) Adds happy + error tests for the 4 invitation control-plane routes: POST /invitation/new, GET /invitation/info, POST /invitation/update, POST /invitation/delete. Patches _user_has_admin_privileges / _user_has_admin_view to avoid extensive get_user_object mocking. * test(proxy/proxy_server): pin config CRUD routes (PR3, partial) Adds happy + error tests for the 8 config-CRUD control-plane routes: POST /config/update, POST|GET /config/field/update|info, GET /config/list, POST /config/field/delete, POST /config/callback/delete, GET /get/config/callbacks, GET /config/yaml. Attaches litellm_config to mock_prisma per-test. * test(proxy/proxy_server): tighten pin assertions per review - test_routes_misc.py: `b"" in response.content` is trivially true; replace with `len(response.content) > 0` so an empty 405 body trips the gate. - test_routes_login_sso.py: `len(response.content) >= 0` is trivially true; tighten to `> 0`. - test_routes_anthropic_beta.py: replace brittle string-literal checks on the serialized JSON (`'"interval_hours": 12' in payload`) with `json.loads` + dict access so the assertion survives any serializer spacing. - test_routes_config.py: `assert status_code in (404, 500)` was too permissive; the handler re-raises HTTPException(404) verbatim, so pin 404 strictly. --------- Co-authored-by: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com> |
||
|
Yassin Kortam
|
d82eb33a60
|
feat(otel): typed semconv-aligned OpenTelemetry instrumentation (#28909) | ||
|
Mateo Wang
|
581c30f1e8
|
[internal copy of #29089] fix: duplicate claude code traces (#29311) | ||
|
ryan-crabbe-berri
|
c754c560dd
|
fix(proxy): link passthrough success spans to the SERVER root OTEL span (#29315)
* fix(proxy): link passthrough success spans to the SERVER root OTEL span Passthrough requests never wired user_api_key_dict.parent_otel_span into the logging metadata, so on success the litellm_request span orphaned into its own trace and the "Received Proxy Server Request" root span was never ended. Setting it once in _init_kwargs_for_pass_through_endpoint fixes both the non-streaming and streaming paths, since update_environment_variables copies that metadata onto the logging object's model_call_details, which is what the OTEL handler reads. Resolves LIT-3443 * fix(proxy): set passthrough parent span after client metadata merge Greptile flagged that litellm_parent_otel_span was assigned before the _metadata.update() calls that merge request-body metadata, so a client body mirroring the internal key could overwrite the real span with a JSON scalar and null the fix for that request. Move the assignment after the merge and add a regression test that fails on the old ordering. * fix(proxy): also set user_api_key after client metadata merge Per Greptile, user_api_key had the same clobber window as the parent span: a passthrough request body mirroring the key could overwrite the authenticated value in the logged metadata. Move it into the same post-merge block and add a deterministic contract test asserting both internal keys resist client-supplied metadata. |
||
|
Sameer Kankute
|
af17400c38
|
feat(a2a): well-known agent-card discovery + LangGraph Platform mode (#28860)
* feat(a2a): well-known agent-card discovery + LangGraph Platform mode
Adds a registration-time discovery flow so admins can paste an upstream
agent URL, see its skills/capabilities, pick what to expose, and have the
proxy front it with a LiteLLM-shaped agent card.
Backend (new litellm/proxy/a2a/ module):
- fetch_well_known_card walks /.well-known/agent-card.json,
/.well-known/agent.json, /agent.json by default. langgraph_platform
mode hits the canonical path with ?assistant_id=<id> (LangGraph
serves one shared endpoint per deployment).
- merge_agent_card overlays LiteLLM overrides on the upstream card:
drops upstream url, forces protocolVersion=1.0, replaces
securitySchemes with LiteLLMKey bearer, emits supportedInterfaces
pointing at the proxy, filters capabilities to a small allowlist,
strips non-v1.0 fields.
- POST /v1/a2a/discover returns the raw upstream card (admin-only) so
the UI can render skills/capabilities for selection.
- create/update/patch agent endpoints pre-generate the agent_id and
run merge_agent_card before storing, so DB.agent_card_params already
embeds the proxy-fronted URL.
UI (ui/litellm-dashboard):
- New AgentCardDiscovery component with a parent-driven plan:
discovery_mode + params + display URL. For LangGraph the parent
composes (api_base, assistant_id); for pure A2A it uses the url
field. Component hides the manual URL input when the parent drives.
- add_agent_form wires discovery for every non-custom agent type and
overlays the user's selections onto agent_card_params at submit,
fixing the bug where dynamic agent forms ignored discovery picks.
Completion-bridge fixes (paired):
- Add kind: "message" to A2A response messages and unwrap result
so it's a Message directly per spec (matches a2a SDK
SendMessageResponse validation).
- Forward A2A metadata to LangGraph runs via extra_body.metadata.
* fix(a2a): preserve agent url, fix streaming chunk envelope, and protect forwarded metadata
- Streaming chunk: move final out of the message object into the
result envelope per the A2A spec.
- Agent card merge: keep upstream url on the stored card so the
runtime invocation path can locate the upstream backend; the public
well-known endpoint already rewrites this field to the proxy URL
before exposing it to clients.
- Completion bridge: apply A2A forward metadata after merging
litellm_params so an agent-configured extra_body cannot
overwrite the forwarded metadata.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(a2a): fix legacy streaming chunk, agent card test, and metadata merge
- providers/litellm_completion: move 'final' out of the message object
into the result envelope per the A2A spec (matches the bridge fix).
- agent endpoints test: the runtime invocation path now preserves the
top-level 'url' on the stored card, so update the assertion to match.
- completion bridge metadata: when forwarding A2A metadata via
extra_body.metadata, merge into any existing extra_body.metadata
instead of replacing it, so an agent-configured metadata block is
preserved (forward metadata still wins on key conflicts).
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(a2a): remove dead duplicate transformation dir; drop SSRF-prone headers field from /v1/a2a/discover
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(a2a): revert accidental html→index.html rename from afc8b10f
The commit afc8b10f bundled real A2A fixes alongside an unintended
re-introduction of the */index.html layout that
|
||
|
Sameer Kankute
|
83fcacad08
|
fix(mcp): ignore stale ids on key save (#29128)
* fix(mcp): ignore stale server ids during key permission validation Prevent virtual key save failures when object_permission still includes MCP server IDs that were deleted from the registry. The validator now drops stale IDs before team scope checks and adds coverage for stale-ID and active unauthorized-ID behavior. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(mcp): use DB find_many to detect stale server IDs, add tool-permission test Addresses greptile feedback on PR #29128: P1 – Race-condition / authoritative-signal concern: Extract _get_stale_mcp_server_ids() helper that performs a single DB find_many (LiteLLM_MCPServerTable) when prisma_client is available, making server-existence checks authoritative even during in-memory registry warm-up. Falls back to the in-memory registry only when no DB client is present (config-only deployments / unit tests). Pass prisma_client through both call sites in key_management_endpoints. P2 – Missing test coverage for mcp_tool_permissions: Add test_validate_stale_ids_in_mcp_tool_permissions_silently_dropped to confirm stale server IDs that appear only as keys in mcp_tool_permissions are also stripped without raising 403. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(mcp): also check in-memory registry when DB is present to avoid misclassifying config-file servers as stale Config-file MCP servers are never written to LiteLLM_MCPServerTable, so a DB-only query would classify them as stale and silently drop their IDs from the authorization check, allowing unauthorized key access. Fix: treat a server ID as stale only when it is absent from BOTH the DB and the in-memory registry (which holds config-file servers). Co-authored-by: Cursor <cursoragent@cursor.com> * fix(mcp): normalize server aliases on key save Co-authored-by: Sameer Kankute <Sameerlite@users.noreply.github.com> * fix: remove unused stale MCP helper and capture normalized object_permission on key generation Co-authored-by: Yassin Kortam <yassin@berri.ai> * test(mcp): keep validate_key_mcp_servers_against_team stub in sync with return contract validate_key_mcp_servers_against_team now returns the (normalized) object_permission, and the key-generation helper assigns that return value back into the request data. The two key-generation tests stubbed the function with a bare AsyncMock, whose default MagicMock return value clobbered object_permission and skipped permission-record creation. Make the stubs pass object_permission through unchanged. * fix(mcp): preserve provided object_permission fields on key update The key-update path reconstructs data.object_permission from a full model_dump(), which marks every field as set. Downstream model_dump(exclude_unset=True) then emits models/blocked_tools/ search_tools as None, and those are non-nullable array columns, so the Prisma write fails whenever the UI submits object_permission with only a subset of fields populated (e.g. a TPM/RPM-only edit). Build the dict with exclude_unset=True so the normalized object_permission keeps the caller's original field set. --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Sameer Kankute <Sameerlite@users.noreply.github.com> Co-authored-by: Yassin Kortam <yassin@berri.ai> Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> |
||
|
Sameer Kankute
|
1d9095f914
|
fix(bedrock): support tool search results + chat annotations (#29120)
* Fix overiding of fastapi_response headers * fix(bedrock): support tool search results and surface citations as annotations Add an optional tool-message search_results path that maps directly to Bedrock toolResult.searchResult blocks, and convert Converse citationsContent into chat completion annotations for user-facing citation metadata. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(format): align bedrock prompt factory with black Reformat the updated bedrock prompt template conversion file so CI black --check passes. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(bedrock): harden citations, search_results mapping, and token counting Resolve mypy issues in citation parsing, only attach url_citation annotations when citation text is stitched into content, fall back to tool content when search_results is empty, and count search_results text in token/TPM preflight paths. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(bedrock): extract tool result helpers to satisfy PLR0915 Refactor _convert_to_bedrock_tool_call_result into smaller helpers so lint passes without changing Bedrock tool result behavior. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(bedrock): count all forwarded search_results fields in token estimates Include source, title, content text, and citations when estimating tokens so large metadata cannot bypass TPM preflight checks. Reformat factory.py with black. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(managed-files): skip content blocks without a type key in get_file_ids_from_messages * fix(bedrock): stitch citations for any punctuation-only text block * fix(bedrock): map null citation source/title to empty annotation strings * fix(bedrock): advance citation offset for text-only citationsContent blocks * fix(bedrock): complete citation TypedDicts for grounding annotations --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> |
||
|
Sameer Kankute
|
c792df64d2
|
feat(mcp): support stateless and stateful clients via session-id routing (#26857)
* feat(mcp): support stateless and stateful clients via session-id routing
- Add session_manager_stateful (stateless=False) alongside stateless
- Route by mcp-session-id: has ID → stateful, initialize (no ID) → stateful, else → stateless
- Peek POST body to detect initialize for routing; replay via wrapped receive
- Handle stale session IDs for both managers
- Add test_mcp_routing_initialize_to_stateful_no_session_to_stateless
- Update test_valid_mcp_session_id_is_preserved, test_concurrent_initialize_session_managers
Made-with: Cursor
* fix(mcp): respect stateful routing and harden initialize detection
Ensure streamable MCP requests are dispatched via the computed target session manager, and guard initialize detection against non-object JSON bodies. Update stale-session test patches to target the stateful manager so routing assertions remain correct.
Made-with: Cursor
* test(mcp): patch stateless/stateful managers in concurrency init test
Update concurrent session-manager initialization test to patch session_manager_stateless and session_manager_stateful directly, matching initialize_session_managers() behavior and preventing NameError from undefined mocks.
Made-with: Cursor
* Fix tests
* Fix tests
* Fix MCP stateful routing edge cases
* Fix stateful MCP auth context refresh
* Fix MCP stateful session cleanup
* fix(mcp): bind stateful sessions to creator and reject hijacks
Stateful mcp-session-id was usable by any authenticated proxy caller. Track
the session creator's hashed API key (or user_id) when a new session is
issued and reject mismatched callers with 403 before _set_or_update_auth_context
overwrites the stored MCPAuthenticatedUser. Also formats nested with-statements
in test_mcp_stale_session.py and fixes a pre-existing AsyncMock mismatch in
test_stale_mcp_session_id_is_stripped.
* fix(mcp): serialize concurrent requests on same stateful session
Bugbot's 'Concurrent requests share context' finding: _update_auth_context
mutates the single MCPAuthenticatedUser stored per session in place on
every request, so two requests sharing one mcp-session-id can overwrite
each other's mcp_servers / auth headers / oauth state / client_ip while
in-flight callbacks are still reading the same object.
Owner-binding alone narrows this to same-principal racing, but the
in-place mutation race remains. Add a per-session asyncio.Lock around
handle_request so concurrent same-session requests run sequentially. The
lock is allocated on demand and torn down with the rest of the session
state on DELETE / idle expiry.
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* fix(mcp): include OAuth2 bearer in stateful session owner fingerprint
UserAPIKeyAuth() for OAuth2 passthrough has no api_key/user_id, so every
OAuth caller fingerprinted to "anonymous" and could hijack another OAuth
caller's mcp-session-id. Hash the upstream Authorization header into the
fingerprint as oauth:<sha256>.
* fix(mcp): don't hold stateful session lock for streaming GETs
The per-session lock wraps handle_request, so a long-lived GET (SSE
stream held open for the life of the session) would block every
subsequent POST on the same mcp-session-id. Only POST/DELETE mutate the
shared MCPAuthenticatedUser, so it's sufficient to serialize those —
GETs run lock-free and stream concurrently.
* fix(mcp): allow None user_api_key_auth in MCPAuthenticatedUser
The set_auth_context / _set_or_update_auth_context / _update_auth_context
helpers in server.py all accept Optional[UserAPIKeyAuth] and pass it
straight into MCPAuthenticatedUser, but the dataclass-style constructor
typed user_api_key_auth as required UserAPIKeyAuth. Mypy flagged this on
the stateful-routing branch:
server.py:3227: error: Incompatible types in assignment (expression
has type "UserAPIKeyAuth | None", variable has type "UserAPIKeyAuth")
server.py:3255: error: Argument "user_api_key_auth" to
"MCPAuthenticatedUser" has incompatible type "UserAPIKeyAuth | None";
expected "UserAPIKeyAuth"
Widen the parameter type to Optional[UserAPIKeyAuth] to match the call
sites. Runtime behavior is unchanged.
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* style: replace with new alias
* fix(mcp): fall back to client_ip in stateful session owner fingerprint
Addresses Greptile review on PR #26857: when no API key, user_id, or
OAuth bearer is available (e.g. unauthenticated/passthrough callers),
the owner fingerprint collapsed to a single 'anonymous' value, allowing
two unrelated callers to drive each other's stateful MCP sessions.
Fold client IP into the fingerprint as a fallback identity signal so
distinct anonymous sources do not share an owner identity.
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* Fix active stateful MCP session cleanup
* test(mcp): cancel leaked stateful auth-context cleanup task
initialize_session_managers() spawns a real asyncio.create_task running
_cleanup_expired_stateful_session_auth_contexts(). The
test_concurrent_initialize_session_managers test was saving and
restoring the session-manager context-manager globals but did not save,
cancel, or restore _stateful_auth_context_cleanup_task.
Because pyproject.toml sets asyncio_default_fixture_loop_scope=session,
the event loop is shared across tests in the same session, so the
leaked task kept running against module-level dicts for the rest of the
test run. Save and cancel the task in the finally block so the test
fully cleans up after itself.
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* Fix stateful MCP session fingerprinting
* Hash MCP session user owner fingerprints
* Fix stale MCP session DELETE cleanup
* fix(mcp): harden owner fingerprint hashing for non-str api keys
_owner_fingerprint_for assumed api_key/user_id supported .encode();
MagicMock-based tests (and any non-str truthy values) crashed with
TypeError before routing. Only hash str/bytes secrets; fall through
otherwise so MCP routing and session tests behave correctly.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Fix MCP stateful cleanup loop resilience
* Fix stateful MCP initialize auth capture
* fix(mcp): drop orphan per-session lock when auth context absent
Defensive cleanup for _stateful_session_locks entries created on
sessions that never enter _stateful_session_auth_contexts. The
periodic cleanup loop only iterates auth_context_last_seen, so such
locks would otherwise live forever. Add a test that reproduces the
leak and verifies the request finalizer pops the lock.
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* chore(mcp): trim verbose comment on lock cleanup
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* Fix stateful MCP delete failure tracking
* fix test
* fix(mcp): cap routing-peek body size to bound pre-dispatch memory
Authenticated clients that POST without an mcp-session-id forced the proxy
to buffer the entire request body before routing, since the peek loop
drained every body chunk to decide whether the JSON-RPC method was
'initialize'. Cap the peek at 4 KB (more than enough for an initialize
envelope) and let the remainder stream through wrapped_receive into the
downstream handler.
* test: replace dall-e-3 with gpt-image-1 in health check and router tests (#27813)
OpenAI returns 'The model dall-e-3 does not exist' for the test account,
breaking test_openai_img_gen_health_check and test_image_generation.
Switch to gpt-image-1, matching the existing TestOpenAIGPTImage1 pattern.
* fix(tests): drop dall-e-only test classes; route live image tests via gpt-image-1
Second wave of failures from the 2026-05-12 DALL-E shutdown:
- tests/image_gen_tests/test_image_edits.py::TestOpenAIImageEditDallE2
and tests/image_gen_tests/test_image_generation.py::TestOpenAIDalle3
are explicitly named for the deprecated models and can't pass; remove.
gpt-image-1 coverage already exists in sibling classes.
- tests/local_testing/test_router.py image gen tests use dall-e-3 only
as a routing example; swap to gpt-image-1.
- tests/local_testing/test_custom_callback_input.py image_generation
success/failure paths swapped to gpt-image-1.
* Fix MCP initialize session active tracking
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* Fix MCP reinitialize session tracking
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* Fix MCP reinitialize auth context aliasing
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* Apply black formatting after merge
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* Run owner-binding 403 before consuming POST body
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* Harden MCP routing peek bound and stateful purge race
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* Remove inadvertently committed Next.js build artifacts
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* Run owner check before stale MCP session cleanup
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* fix(mcp): reverse cleanup ordering to terminate transport before clearing owner
Reverses _purge_expired_stateful_session_auth_contexts so the transport
is popped from server_instances and terminated BEFORE owner/auth tracking
is cleared. The previous order left a window where _stateful_session_owners
was already empty but server_instances still served the session, so a
concurrent request would observe expected_owner is None and bypass the
owner-binding check. Addresses Greptile review on PR #26857.
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
* test(mcp): fully reset stateful session tracking in auth-context refresh test
Use _remove_stateful_session_tracking in teardown so the test no longer
leaks _stateful_session_auth_context_last_seen and _stateful_session_locks
between tests, matching the cleanup used by the sibling stateful tests.
* fix(mcp): cap concurrent stateful sessions per caller to bound memory
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Co-authored-by: Mateo Wang <mateo-berri@users.noreply.github.com>
Co-authored-by: Sameerlite <sameerlite@users.noreply.github.com>
Co-authored-by: yuneng-jiang <yuneng@berri.ai>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
Co-authored-by: Claude Babysitter <claude@anthropic.com>
Co-authored-by: mateo-berri <mateo@berri.ai>
|
||
|
Sameer Kankute
|
70d2748d80
|
fix(proxy): map stripped batch body.model to proxy alias for auth (#29264)
* fix(proxy): map stripped batch body.model to proxy alias for auth replace_model_in_jsonl rewrites JSONL body.model to the provider id before upload; batch file access checks must resolve that id back to model_name so keys granted the proxy alias are not rejected with 403. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(proxy): surface resolved proxy alias in batch file 403 detail --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> |
||
|
Sameer Kankute
|
0ffab87da4
|
fix(router): enforce deployment budgets for dynamically added models (#29273)
* fix(router): enforce deployment budgets for dynamically added models Register deployment max_budget/budget_duration when models are added via upsert_deployment (e.g. /model/new) so RouterBudgetLimiting matches startup model_list behavior. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(router): address CI lint and router coverage for budget sync helpers Remove unused RouterBudgetLimiting import and add router unit tests for deployment budget helper methods required by router_code_coverage. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(router): clear stale deployment budget on upsert without limits Unregister deployment budget config when max_budget/budget_duration are removed, including upsert replace paths. Hoist provider budget logger lookup outside the provider loop. Co-authored-by: Cursor <cursoragent@cursor.com> * Fix mypy * fix black --------- Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
ryan-crabbe-berri
|
909a5f597a
|
fix(mcp): resolve key.access_group_ids → MCP servers (ungated) (#29195)
* fix(mcp): resolve key.access_group_ids → MCP servers (ungated) A teamless virtual key whose unified access_group_ids grant an MCP-granting access group now sees and can call that server instead of getting an empty list / 403. The key path previously read only the legacy object_permission; this folds key.access_group_ids into the key's base scope (ungated), mirroring can_key_call_model's fallback. The gated assigned_*-checked override is unchanged. * fix(mcp): expand name/alias entries in key access-group servers access_mcp_server_ids may hold server names/aliases, not just ids. The new ungated key path now runs them through expand_permission_list at the source, so the early-return and union branches both surface resolved ids — matching the legacy object_permission path and the gated extras path. |
||
|
Mateo Wang
|
a55817cbc6
|
fix(anthropic): stop injecting unsupported output_config.effort=xhigh for Claude Code on Sonnet/Opus 4.6 (#29304)
* fix(anthropic): don't inject output_config.effort=xhigh on models without xhigh The legacy-thinking translator on the /v1/messages route mapped any thinking.budget_tokens >= 24000 to effort=xhigh and injected it into output_config without checking model support. Claude Code's default thinking budget (31999) hit this bucket, so Sonnet 4.6 (and Opus 4.6) on Bedrock/Vertex started returning 400 output_config.effort: Input should be 'low', 'medium', 'high' or 'max' Gate the xhigh choice on _supports_effort_level(model, "xhigh"), the same capability check the reasoning_effort path already uses. Models that advertise xhigh (Opus 4.7) keep it; everything else falls to high. Fixes #29282 * test(anthropic): pin Opus 4.6 in legacy-thinking xhigh-clamp regression test Opus 4.6 (bare, bedrock/invoke, vertex_ai) has supports_adaptive_thinking but no supports_xhigh_reasoning_effort, so it hits the same clamping path as Sonnet 4.6. It was named in the PR scope but lacked a pinned regression guard; add the three variants to the parametrize list. |
||
|
michelligabriele
|
68852ef165
|
fix(teams): expose keys_count on /v2/team/list and wire UI Resources badge (#28502) | ||
|
ryan-crabbe-berri
|
9918a9c78c
|
fix(guardrails): persist disable_global_guardrails on keys (#29233)
* fix(guardrails): restore disable_global_guardrails persistence for keys The per-key/team "Disable Global Guardrails" toggle silently stopped working after #17042, which removed `disable_global_guardrails` from the key/team request models and from the premium metadata allowlist. Without those, the UI's top-level field was dropped by pydantic and never folded into key `metadata`, so the runtime gate always read False and global default_on guardrails kept running. Restore the request-model fields (KeyRequestBase, NewTeamRequest, UpdateTeamRequest) and the `LiteLLM_ManagementEndpoint_MetadataFields_Premium` entry so the flag is promoted into metadata again. Because the key edit form always submits the flag (false by default), guard the UI so it is only sent when it actually changed (edit) or is enabled (create) — this keeps the premium gate on enabling intact while not 403-ing non-premium users who edit unrelated key fields, mirroring how guardrails/tags are already stripped. * test(guardrails): cover disable_global_guardrails toggle-off + clarify premium field comment Add a prepare_metadata_fields case asserting `disable_global_guardrails: False` overwrites an existing `True`, and rewrite the PREMIUM_METADATA_FIELDS comment to explain why boolean premium fields are excluded from the empty-value strip loop. |
||
|
Mateo Wang
|
bae04591b2
|
feat(anthropic): add Claude Opus 4.8 and prune reasoning-effort flags (#29238)
* feat(anthropic): add Claude Opus 4.8 and prune reasoning-effort flags Register claude-opus-4-8 across the anthropic/bedrock/vertex/azure cost-map entries, BEDROCK_CONVERSE_MODELS, and the setup-wizard provider list. Prune two reasoning-effort fields from the cost map: - Drop supports_minimal_reasoning_effort from the Claude fleet (58 entries). "minimal" is not a real Anthropic effort level (the API accepts only low/medium/high/xhigh/max), so LiteLLM degrades it to "low" regardless; the flag was inert and misleading on Anthropic. - Remove tool_use_system_prompt_tokens everywhere (103 entries). It is not in the ModelInfo type and is read by no production code. Update the affected config/schema tests; the reasoning-effort registry tests now assert the Claude fleet omits supports_minimal. * fix(anthropic): recognize output_config effort after minimal-flag prune Pruning supports_minimal_reasoning_effort from the Claude fleet removed the only "supports effort param" marker from 11 Opus 4.5 / mythos-preview map entries that lack supports_output_config. _model_supports_effort_param then returned False for them, so output_config was wrongly dropped under drop_params=True -- regressing test_anthropic_model_supports_effort_param_recognizes_supporting_models for claude-opus-4-5-20251101 and the mythos preview. - _model_supports_effort_param now treats supports_output_config as a sufficient signal, matching the bedrock-invoke call sites that already check supports_output_config OR a reasoning-effort flag. Shared map lookup extracted into _supports_model_capability. - Add supports_output_config: true to the 11 Opus 4.5 / mythos entries that lost their only marker, restoring prior effort-forwarding behavior without re-adding the inert minimal flag. |
||
|
yuneng-jiang
|
76f56c3283
|
fix(tests/vcr): mint Google OAuth tokens live to prevent stale-token replay (#29229)
The Redis-backed VCR layer was recording and replaying the Google OAuth2/STS token-mint call. The replayed ya29.* access token is long-expired, but its recorded expires_in keeps credentials.expired False, so litellm never refreshes it and sends the stale token to a live Vertex/Gemini endpoint, which returns 401 ACCESS_TOKEN_EXPIRED. This broke live partner-model tests whose completion call is not itself cassette-backed (e.g. test_vertex_ai_llama_tool_calling). Force credential-exchange hosts to pass through live (never recorded, never replayed) by returning None from before_record_request, mirroring the existing telemetry passthrough, so a fresh token is minted each run. Regression from #28826, which added OAuth-token matcher tolerance plus TTL-refresh-on-read so a stale token episode matched and never expired. |
||
|
yuneng-jiang
|
5e2d75d75d
|
bump deps (#29208) (#29226)
* fix(deps): bump vulnerable proxy dependencies (starlette/fastapi, granian, pyarrow, semantic-router) Resolve known CVEs flagged by osv-scanner/grype against uv.lock. All bumped versions verified to resolve, install, and pass the proxy auth/route/middleware unit suites (717 tests) plus an import smoke on the new stack. - starlette 0.50.0 -> 1.1.0 (CVE-2026-48710 "BadHost", GHSA-86qp-5c8j-p5mr): versions <1.0.1 reconstruct request.url from the unvalidated Host header, poisoning request.url.path. Required raising fastapi 0.124.4 -> 0.136.3, which dropped fastapi's starlette<0.51.0 cap; an explicit starlette>=1.0.1 floor blocks regression to a vulnerable transitive resolution. The proxy's own auth already reads scope["path"] via get_request_route, but the locked starlette still flagged in container scanners and left other request.url consumers exposed. - granian 2.5.7 -> 2.7.4 (CVE-2026-42544, unauthenticated DoS via WebSocket subprotocol header panic; CVE-2026-42545, WSGI response-header-panic DoS). granian is a selectable proxy server (proxy_cli). - pyarrow 22.0.0 -> 23.0.1 (CVE-2026-25087 / PYSEC-2026-113). - semantic-router 0.1.12 -> 0.1.15: 0.1.12 was yanked (CVE-2026-42208 — its unbounded litellm pin could resolve a credential-exfiltrating litellm==1.82.8 wheel). Not fixable by bump: diskcache 5.6.3 (CVE-2025-69872, unsafe pickle deserialization) has no upstream fix and is left pinned; exploiting it requires write access to the local cache directory. Relock side effect: sse-starlette 3.4.2 -> 3.4.4. * deps: relax exact pins in optional extras to compatible ranges The proxy/optional extras exact-pinned every dependency, which (1) forces downstream `pip install litellm[proxy]` consumers into version lockstep and (2) blocks them from pulling transitive security patches without forking — the structural cause behind needing a litellm release to clear the starlette CVE in the previous commit. Convert the ordinary extras deps to `>=current,<next_major` ranges, mirroring the core [project].dependencies style. Reproducibility for litellm's own Docker/CI is unaffected: images install via `uv sync --frozen`, and the lock re-resolves to the identical versions (no locked version changed). Kept exact-pinned: - litellm-proxy-extras, litellm-enterprise — litellm's own sub-packages, versioned in lockstep with the release. - opentelemetry-api/sdk/exporter-otlp — must resolve to matching versions. - grpcio — supply-chain-pinned to a vetted, aged release. Also corrects the stale comment claiming the extras are exact-pinned for Docker reproducibility (the images use the lock, not these pins). * fix(ci): resolve license-check lookup version from the floor for ranged deps check_licenses.py derived the PyPI lookup version with `next(iter(req.specifier))`, which returns an arbitrary specifier clause. For a range like `>=0.12.1,<1.0` it picked the upper bound (`1.0`) — a version that doesn't exist on PyPI — so the license lookup 404'd and the package was flagged as having an unknown license. The previous commit's switch from exact pins to ranges exposed this for soundfile, pyroscope-io, redisvl, diskcache, and mlflow (the ranged deps not already in liccheck.ini's allowlist). Prefer a lower-bound/exact version (a real released version) for the lookup. * fix(proxy): set strict_content_type=False on the FastAPI app Starlette 1.0 / FastAPI 0.13x flipped the default to strict_content_type=True, which refuses to parse a JSON request body when the client omits the Content-Type header. The proxy previously accepted those requests, so the fastapi/starlette bump in this PR would silently break clients that don't send a Content-Type. Restore the prior lenient behavior explicitly. Co-authored-by: stuxf <70670632+stuxf@users.noreply.github.com> |
||
|
michelligabriele
|
928f09f8a4
|
fix(datadog): drain cost-management queue + opt-in FinOps tag allowlist (#28487)
* fix(datadog): drain cost-management queue + opt-in FinOps tag allowlist * fix(datadog): guard non-dict callback_specific_params + log empty aggregation * fix(datadog): block user-controlled tags from overwriting reserved cost-attribution dimensions * fix(datadog): cast metadata to dict[str, Any] to satisfy mypy |
||
|
Sameer Kankute
|
69afcd09d0
|
fix(vertex-ai): use DB credentials in video handlers + implement Veo video edit (#29098)
* fix(vertex-ai): pass litellm_params to validate_environment in video handlers and implement video edit for Veo - Pass litellm_params to validate_environment in 11 video handler call sites (remix, create_character, get_character, edit, extension, delete) so DB-stored Vertex AI credentials are used instead of falling back to ADC - Implement transform_video_edit_request/response for VertexAI: fetches source video via fetchPredictOperation then submits a new predictLongRunning request with the video bytes/gcsUri + edit prompt Co-authored-by: Cursor <cursoragent@cursor.com> * fix(vertex-ai): hoist fetchPredictOperation into handlers to avoid blocking event loop - Add get_video_edit_prefetch_params() to BaseVideoConfig (returns None) - VertexAI overrides it to return the fetchPredictOperation URL/body - Both sync and async video_edit handlers call this and use their shared httpx client for the fetch, passing the result as prefetched_source_data - transform_video_edit_request is now a pure transform with no HTTP calls - Fix extra_body.pop() mutation by working on a shallow copy Co-authored-by: Cursor <cursoragent@cursor.com> * fix(vertex-ai): include prefetch call inside _handle_error try/except block Co-authored-by: Cursor <cursoragent@cursor.com> * fix(videos): add prefetched_source_data param to all transform_video_edit_request overrides Co-authored-by: Cursor <cursoragent@cursor.com> * fix(video_edit): keep transform/pre_call outside try so validation errors propagate Move transform_video_edit_request and logging_obj.pre_call outside the try/except that wraps HTTP calls in (async_)video_edit_handler so that ValueError validation errors (e.g. 'source video not complete yet') are not silently wrapped as 500s by _handle_error. The prefetch HTTP call keeps its own try/except so its errors are still mapped through the provider's error handler. Matches the pattern used by video_extension_handler and video_remix_handler. Co-authored-by: Yassin Kortam <yassin@berri.ai> * refactor(vertex_ai): delegate get_video_edit_prefetch_params to status retrieve Co-authored-by: Yassin Kortam <yassin@berri.ai> * Fix varia review * fix(video_edit): route transform errors through _handle_error Wrap transform_video_edit_request and pre_call in the same try/except as the HTTP call in sync and async handlers so validation failures (e.g. source video not complete) return typed LiteLLM exceptions. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Yassin Kortam <yassin@berri.ai> |
||
|
Sameer Kankute
|
eef1ec3e8d
|
fix(proxy): enforce tag budgets for key-level tags (#29108)
* fix(proxy): enforce tag budgets for key-level tags Merge API key metadata.tags into request_data before _tag_max_budget_check so per-tag budgets apply when tags are set on the key at creation time. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(auth): avoid false reject for key-inherited tags Run reject_clientside_metadata_tags before key-tag injection, then inject key metadata tags immediately before tag budget checks so key tags still enforce budgets without being treated as client-supplied tags. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
Yassin Kortam
|
d5d6b26a72
|
fix: improve bedrock streaming hot path perf (#28720) | ||
|
Sameer Kankute
|
5bd59b33e6
|
feat(guardrails): wire apply_guardrail into proxy logging callbacks (#28970)
* feat(guardrails): wire apply_guardrail into proxy logging callbacks Route /apply_guardrail through pre/post proxy hooks and LiteLLM success/failure handlers so Langfuse and OTEL integrations receive input/output on guardrail-only requests. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(guardrails): fix Greptile review comments on apply_guardrail logging Co-authored-by: Cursor <cursoragent@cursor.com> * fix(apply_guardrail): preserve original exception and capture modified response - Capture return value from post_call_success_hook so callback-modified responses propagate to the caller. - Wrap success/failure logging calls in defensive try/except so logging infrastructure failures don't replace the user-visible response or mask the original guardrail exception. Co-authored-by: Yassin Kortam <yassin@berri.ai> * Fix mypy * fix(apply_guardrail): isolate failure logging and use post-hook response for logging - Split async_failure_handler and post_call_failure_hook into independent try/except blocks so a callback bug in one does not silently skip the other. - Build response_for_logging inside _emit_guardrail_success_logs after post_call_success_hook runs, so logged data matches the response the caller actually receives when the hook modifies the response. Co-authored-by: Yassin Kortam <yassin@berri.ai> * fix(apply_guardrail): fix black formatting and update tests for fastapi_request param - Run black on guardrail_endpoints.py to fix CI formatting check - Add _mock_proxy_logging() helper to enterprise guardrail tests to patch proxy-server globals imported at call time - Pass fastapi_request=Mock() in all direct apply_guardrail test calls to match updated function signature Co-authored-by: Cursor <cursoragent@cursor.com> * fix(guardrails): use transformed exception from post_call_failure_hook in apply_guardrail Co-authored-by: Yassin Kortam <yassin@berri.ai> * fix(guardrails): isolate sync/async logging handlers in apply_guardrail Separate each logging handler call into its own try/except so a failure in the async handler does not silently skip the sync handler submission (and vice versa). Matches the docstring's defensive intent. Co-authored-by: Yassin Kortam <yassin@berri.ai> * fix(apply_guardrail): guard transformed_exception with isinstance check Co-authored-by: Cursor <cursoragent@cursor.com> * test(guardrails): mock proxy globals in not_found test and share apply_guardrail logging fixture - Add proxy-server global mocks to test_apply_guardrail_not_found so the failure-path post_call_failure_hook call doesn't touch the real proxy logging singleton. - Extract the duplicated _mock_proxy_logging context manager out of the two enterprise apply_guardrail test files into a shared conftest fixture so the helper stays in one place. * fix(guardrails): use update_messages to keep logging obj in sync Co-authored-by: Yassin Kortam <yassin@berri.ai> --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Yassin Kortam <yassin@berri.ai> Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> |
||
|
Mateo Wang
|
95015de733
|
feat: add support for claude code goal mode for bedrock opus output config (#28898)
* feat: support goal mode for claude on bedrock
* fix failing lint test
* addressing greptile comments
* fixing failed test
* address greptile: copy output_config and warn on dropped converse format
* fix(bedrock): skip redundant output_config normalization on Converse reasoning_effort path
When reasoning_effort is mapped via _handle_reasoning_effort_parameter, the
resulting output_config is already normalized via
normalize_bedrock_opus_output_config_effort. Mark it as normalized so
_prepare_request_params can skip the redundant call (and the associated
get_model_info lookup) on every request.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* test(reasoning-effort-grid): reflect Bedrock opus-4-6 xhigh→max clamping
* fix(bedrock): stop leaking output_config marker and message-content mutation
* fix(bedrock): guard effort key access in normalize_bedrock_opus_output_config_effort
Defensively check that 'effort' is a valid key in _BEDROCK_OUTPUT_CONFIG_EFFORT_ORDER
before indexing, to prevent a KeyError if the hardcoded guard tuple ever drifts from
the order dict's keys.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(bedrock): drop dead second clause in effort normalization guard
The 'effort not in _BEDROCK_OUTPUT_CONFIG_EFFORT_ORDER' check is
unreachable once 'effort not in ("xhigh", "max")' has been ruled out,
since both literals are present in the order dict. Keep the literal
membership check and let the dict lookups below speak for themselves.
* fix(bedrock): clamp output_config.effort against ceiling for any known value
The early return when effort was not 'xhigh'/'max' meant a ceiling of
'low' or 'medium' would silently forward an out-of-range value. Gate on
the known effort ordering instead so the ceiling comparison runs for
every recognized effort.
* test(grid_spec): use _CAPS_OPUS_4_7 for non-Bedrock opus-4-6 entries
claude-opus-4-6 now declares supports_xhigh_reasoning_effort in the model
map, so production accepts xhigh on Azure AI and Vertex AI routes. Update
those grid_spec entries to match production capabilities so expected()
predicts 200 for xhigh instead of 400.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* test(grid_spec): revert xhigh caps for non-Bedrock opus-4-6
azure_ai/claude-opus-4-6 and vertex_ai/claude-opus-4-6 do not declare
supports_xhigh_reasoning_effort in model_prices_and_context_window.json.
Azure AI upstream rejects xhigh with HTTP 400 ("Supported levels: high,
low, max, medium"). Restore _CAPS_4_6 so the grid predicts 400 for
xhigh, matching production capabilities.
* fix: stop advertising xhigh effort on Opus 4.5/4.6
Only Opus 4.7 supports the xhigh reasoning effort level. Remove the
supports_xhigh_reasoning_effort flag from every Opus 4.5 and Opus 4.6
entry (direct Anthropic, Bedrock, and regional variants) in both model
catalog files.
On the direct Anthropic path there is no effort clamp, so flagging 4.5/4.6
as xhigh-capable caused litellm to forward xhigh to a model that rejects it
(and made get_model_info misreport the capability). xhigh now correctly
degrades to high / raises on those models.
Bedrock graceful degradation for Claude Code goal mode is unaffected: it
relies solely on the bedrock_output_config_effort_ceiling clamp (4.5->high,
4.6->max, 4.7->xhigh), which runs before validation, so xhigh requests to
older Bedrock Opus models are still silently lowered rather than rejected.
Update effort-gating tests to reflect that 4.5/4.6 no longer accept xhigh.
* fix: clamp xhigh effort on Bedrock Invoke /v1/messages instead of rejecting
Claude Code "goal mode" sends output_config.effort=xhigh over the Anthropic
/v1/messages API, which routes Bedrock models through
AmazonAnthropicClaudeMessagesConfig. That path validated effort against the
model's native capability and raised 400 for xhigh on Opus 4.6, while the
chat-completions paths (Converse + Invoke) already clamp xhigh to the model's
bedrock_output_config_effort_ceiling. That asymmetry broke goal mode on the
exact API surface Claude Code uses.
Apply the same ceiling clamp on the messages path before the shared effort
gate runs, so xhigh degrades to max on Opus 4.6 (and stays xhigh on 4.7).
Scoped to adaptive-thinking models and to models that declare a ceiling, so
Sonnet 4.6 (no ceiling) and Opus 4.5 (budget mode) are unaffected and still
reject xhigh.
* fix(bedrock): preserve user output_config when applying reasoning_effort
- Converse path: merge mapped effort into existing output_config via
setdefault instead of overwriting it, matching the Anthropic Messages
path. Prevents user-supplied output_config.format from being silently
dropped when reasoning_effort is also provided.
- tests: clear _get_local_model_cost_map lru_cache in the autouse
fixture alongside get_bedrock_response_stream_shape to avoid stale
cache leakage between tests.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(bedrock): pre-clamp reasoning_effort for chat invoke; correct test caps
- Add _clamp_adaptive_reasoning_effort_for_bedrock to AmazonAnthropicClaudeConfig
so raw reasoning_effort=xhigh degrades to the model's bedrock effort ceiling
before AnthropicConfig.map_openai_params converts it to output_config.
Mirrors converse path (_handle_reasoning_effort_parameter) and messages path
(_clamp_adaptive_reasoning_effort_for_bedrock) so the three Bedrock paths
are consistent.
- grid_spec: restore caps=_CAPS_4_6 for Bedrock converse/invoke Opus 4.6 entries
so the test reflects the model's actual JSON capabilities. Teach expected()
to bypass the xhigh/max cap check when bedrock_effort_ceiling will clamp
the wire effort, so the test still passes for Bedrock's graceful degradation
contract without lying about native model caps.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
---------
Co-authored-by: Dennis Henry <dennis.henry@okta.com>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
|
||
|
Sameer Kankute
|
157e7a0f20
|
fix(containers): record ownership for service-account keys + fix Prisma Json serialization (#28990)
* fix(containers): record ownership for service-account keys + fix Prisma Json field serialization - Track containers created implicitly via /v1/responses by extracting container IDs from the response output and calling record_container_owner for each one, so subsequent file-API calls from the same service account pass ownership checks. - Fix DataError: Prisma Python requires Json fields to be JSON strings; serialize file_object with json.dumps() before insert/update in LiteLLM_ManagedObjectTable. - Add collect_container_ids_from_responses_response utility to responses/utils.py that walks all output item shapes (code_interpreter_call, message annotations). - Tests: two new cases covering the responses-tracking path and the end-to-end record-then-assert flow for service accounts with team scope. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(containers): swallow all exceptions in ownership hook; tighten file_object_json type to str Co-authored-by: Cursor <cursoragent@cursor.com> * fix(containers): parse file_object JSON string in existing ownership test Co-authored-by: Cursor <cursoragent@cursor.com> * fix: container ownership recording bugs - Remove unreachable _aresponses_websocket from route_type set in base_process_llm_request; the WebSocket endpoint never flows through base_process_llm_request, so this branch was dead code that gave a false impression of coverage. - Drop the HTTPException re-raise in record_container_owners_from_responses_response so per-container failures (including HTTP 403/500 from conflicting ownership rows) no longer abort the batch and skip recording for the remaining container IDs in the same response. Co-authored-by: Yassin Kortam <yassin@berri.ai> * fix(containers): record ownership for streaming /v1/responses too Streaming /v1/responses returns through the select_data_generator branch in base_process_llm_request and bypasses the non-streaming ownership tail, so code-interpreter containers created mid-stream were never written to LiteLLM_ManagedObjectTable. Follow-up file API calls would then 403. Wrap the SSE generator so container ownership is recorded once the upstream iterator finishes assembling completed_response. Also covers the background-polling path, which loops body_iterator end-to-end. Co-authored-by: Yassin Kortam <yassin@berri.ai> --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Yassin Kortam <yassin@berri.ai> |
||
|
ryan-crabbe-berri
|
ea015332d8
|
fix(mcp): resolve team.access_group_ids → MCP servers (#28997)
* fix(mcp): resolve team.access_group_ids → MCP servers A virtual key whose team has an MCP-granting access group attached via /v1/access_group now sees that server through /v1/mcp/server (and can call tools on it) instead of getting an empty list. The runtime already resolves the key's unified access_group_ids; this adds the symmetric resolution on the team side, mirroring the model-side pattern in can_team_access_model — the group being on the team is itself the gate, so no assigned_team_ids re-check is needed. Resolves #27657 * chore(mcp): address greptile review on team access-group resolver Forward already-imported prisma_client / user_api_key_cache / proxy_logging_obj to _get_mcp_server_ids_from_access_groups so it skips its lazy re-import path. Update test docstring + assertions to reflect that the resolver is invoked with [] (and short-circuits without DB access) rather than skipped entirely. |
||
|
Sameer Kankute
|
3d0e0cee56
|
[Feat] Add tool calling support for gemini and vertex ai live api (#26590)
* Add tool calling support for gemini and vertex ai live api
* Fix greptile reviews
* Add new functionality behind flag
* fix greptile issues
* Fix greptile review
* Fix greptile review
* Fix greptile review
* Fix greptile review
* Fix greptile review
* fix lint
* fix(realtime): address P1 issues - guardrail timing and inputAudioTranscription default
- Remove early guardrail turn-detection update that consumed first setup slot
- Add inputAudioTranscription default in Gemini deferred-mode setup
- Add tests for both fixes
Made-with: Cursor
* fix(realtime): inject turn_detection into first session.update for deferred mode
- Instead of sending turn_detection as separate message (which gets dropped), inject it into the first client session.update
- This ensures guardrails work correctly in deferred mode
- Add test for turn_detection injection in deferred mode
Made-with: Cursor
* fix(realtime): emit response.created preamble before tool-call events
- Emit response.created, output_item.added, and conversation.item.created for function calls
- Ensures OpenAI Realtime API spec compliance
- Add test for preamble emission
Made-with: Cursor
* fix(realtime): add response.output_item.done to complete tool-call sequence
- Emit response.output_item.done between function_call_arguments.done and conversation.item.created
- Required by OpenAI Realtime spec to finalize function-call items
- Update test to verify complete event sequence
Made-with: Cursor
* fix(realtime): emit response.done after tool-call sequence (P0 CRITICAL)
- Add response.done event after tool-call loop to signal response completion
- Required by OpenAI SDK clients to submit tool results
- Without this, clients stall indefinitely waiting for response completion
- Update test to verify complete 6-event sequence including response.done
Made-with: Cursor
* fix(realtime): include function name in toolResponse (P1)
- Store call_id → name mapping when receiving toolCall from Gemini
- Look up and include name in functionResponses when sending tool results
- Required by Gemini Live API spec for proper tool call routing
- Add test to verify name field is included in round-trip
Made-with: Cursor
* fix: resolve merge conflict markers in UI build chunk
Take litellm_internal_staging version of e1a670efcb966aaa.js after
incomplete merge left conflict markers in the committed artifact.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(vertex_ai/realtime): call super().__init__() to initialize tool call state
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(realtime): correct guardrail flag and event-mapping fallback
- realtime_streaming: only mark _guardrail_turn_detection_update_sent
when the message was actually delivered to the backend. The provider
transformation (e.g. Gemini after initial setup) may silently drop
session.update; previously we set the flag anyway, falsely claiming
the disable was sent and preventing any retry on subsequent
session.created events. _send_to_backend now returns whether at
least one transformed message was sent.
- gemini realtime transformation: avoid shadowing the outer
openai_event variable in map_openai_event's fallback loop. With
the new toolCall entry now last in MAP_GEMINI_FIELD_TO_OPENAI_EVENT,
an unmatched key would otherwise leak FUNCTION_CALL_ARGUMENTS_DONE
and skip the ValueError raise. Use a distinct loop variable so the
is-None check correctly raises for unknown Gemini messages.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini/realtime): reset response IDs after tool-call response.done
After closing a tool-call response, clear current_output_item_id and
current_response_id so post-tool model turns emit a fresh response.created
preamble. Add regression tests and align guardrail turn_detection test with
GA session shape; apply Black formatting.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix lint
* fix(realtime): log injected message and forward guardrail VAD-disable on Gemini
- Move store_input() after the guardrail turn_detection injection in
client_ack_messages so audit logs reflect what is actually forwarded
to the backend (previously the unmodified pre-injection message was
logged).
- In Gemini's _handle_session_update, allow a session.update that only
carries a turn_detection change to be forwarded as a follow-up Gemini
setup with realtimeInputConfig.automaticActivityDetection set, even
after the initial setup. This restores the guardrail layer's ability
to disable VAD auto-response in non-deferred mode (the default Gemini
flow), which was a regression after _handle_session_update started
silently dropping subsequent session.update messages. Both flat
beta-style and nested GA-style turn_detection payloads are accepted.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini/realtime): resolve mypy TypedDict errors in transformation
Align realtime event payloads and setup types with OpenAI/Gemini TypedDicts so mypy passes and tool-call events type-check correctly.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(realtime): forward turn_detection updates for Vertex; respect partial VAD config; cache setup after send
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(realtime): consolidate send-and-cache, guard session.update lookup, preserve client turn_detection in GA remap
- Replace duplicated transform/send/cache logic in client_ack_messages with a call to _send_to_backend so future changes stay in one place.
- VertexAIRealtimeConfig.transform_realtime_request now uses .get('session') or {} for the first session.update so a malformed client payload no longer crashes the connection.
- Move the audio-transcription guardrail turn_detection injection to run BEFORE the beta->GA session remap. This lets the injected create_response ride along with any client-provided turn_detection fields (e.g. silence_duration_ms) into the nested audio.input.turn_detection path produced by the remap instead of being stranded as a separate root-level dict.
- Update the deferred-mode injection test to assert the GA-shaped location.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): pop tool_call_id mapping after use to bound memory
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(realtime): correct deferred-setup session.created modalities and reset IDs after response.done
- Convert provider's real session.created to session.updated when a synthetic
one was already forwarded so clients receive the authoritative modalities
derived from their session.update instead of the synthetic placeholder.
- Reset current_response_id / current_output_item_id after Gemini RESPONSE_DONE
so a toolCall arriving in a later frame starts a fresh response instead of
reusing the completed response's ID and emitting a duplicate response.done.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini-realtime): preserve nested turn_detection through map_openai_params
After the GA remap moves session.turn_detection into session.audio.input.turn_detection,
Gemini's map_openai_params only looks at top-level keys and silently drops it. Normalize
the extracted turn_detection back to the top level on first session.update so the guardrail
create_response:False (and any client-provided VAD settings) reach the Gemini setup.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(realtime): normalize Vertex AI nested turn_detection and unify session.created guardrail ordering
- Vertex AI _build_vertex_ai_setup_config now lifts nested
audio.input.turn_detection to the top level before calling
map_openai_params, mirroring the parent GeminiRealtimeConfig
behavior. Without this, guardrail-injected create_response: False
was silently dropped for GA-protocol Vertex AI clients.
- realtime_streaming session.created handling now sends the
(possibly re-typed) event first and then triggers the guardrail
turn-detection update for both first and duplicate cases, removing
the inconsistent guardrail-then-event ordering for duplicates.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(realtime): tolerate non-dict turn_detection in guardrail injection
When a client sends a session.update whose turn_detection field is None or
a non-dict value (e.g. "auto"), the guardrail injection used setdefault
followed by item assignment on the returned value, raising TypeError. The
inner except only caught JSONDecodeError/AttributeError, so the TypeError
escaped to the outer Exception handler that wraps the entire client_ack
loop, killing the connection. Replace non-dict turn_detection with a
fresh dict carrying create_response=False so the guardrail still applies
without crashing the loop.
* fix(gemini realtime): default synthetic session.created modalities to AUDIO
The synthetic session.created event emitted in deferred setup mode used
TEXT as the default for responseModalities, while _handle_session_update
defaults to AUDIO. Align the default so clients reading modalities from
the initial session.created see the correct value for live sessions.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(vertex_ai/realtime): drop follow-up session.update to avoid 1007 close
Vertex AI Live treats setup as a first-and-only client message; emitting a
second setup with realtimeInputConfig only closes the websocket with a 1007
policy error. Reverting the follow-up-setup branch restores the pre-existing
no-op behavior for subsequent session.update messages.
* fix(gemini realtime): default responseModalities to AUDIO in delta events
Align return_new_content_delta_events with the AUDIO defaults used in
_handle_session_update and transform_session_created_event so deferred
session config does not produce TEXT-typed delta events for audio data.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): default response.done modalities to AUDIO and correct audio-done test
* fix(realtime): set guardrail turn_detection flag only after successful send
Previously the _guardrail_turn_detection_update_sent flag was set inline
during message rewriting in client_ack_messages, before the modified
session.update was forwarded to the backend. If _send_to_backend raised
(e.g. backend WebSocket disconnect), the exception was caught and the
loop continued, but the flag remained True — permanently disabling the
guardrail create_response=False injection for the rest of the session.
Neither the client_ack_messages path nor the
_maybe_send_guardrail_turn_detection_update backup path would retry.
Track the injection locally and only set the flag after _send_to_backend
returns a truthy sent result, matching the pattern used by
_maybe_send_guardrail_turn_detection_update.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(vertex_ai realtime): keep VAD enabled when guardrails inject create_response: False
map_automatic_turn_detection sets disabled=True whenever create_response is
absent OR False. Transcription guardrails inject create_response: False to
suppress auto-responses while expecting VAD to stay active, but the previous
override in _build_vertex_ai_setup_config only fired when create_response was
absent, leaving disabled=True and silently breaking speech detection and
transcription events. Vertex Live has no 'VAD on, no auto-response' mode, so
always keep VAD active in the setup config.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): normalize GA-remapped session fields before mapping
map_openai_params only recognises the flat OpenAI-beta keys (modalities,
input_audio_transcription, turn_detection). For GA clients the upstream
shim renames these into the nested GA schema (output_modalities,
audio.input.transcription, audio.input.turn_detection), causing them to
be silently dropped in _handle_session_update. Add a normalization helper
that surfaces the GA-remapped values back at the top level so the
existing mapping logic picks them up. Without this, a GA client
explicitly requesting modalities=['text'] would still default to audio
output.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(vertex_ai/realtime): normalize all GA-remapped session fields before mapping
Previously _build_vertex_ai_setup_config only lifted nested turn_detection
back to the top level. GA clients' output_modalities and
audio.input.transcription were silently dropped because map_openai_params
only recognises the flat OpenAI-beta keys. Use the parent's
_normalize_session_payload_for_mapping so modalities, transcription, and
turn_detection are all surfaced before mapping.
* fix(realtime): force create_response=False in all client session.update turn_detection when audio guardrails active
Prevents a client from re-enabling Gemini/GA VAD auto-response (and thereby
bypassing the audio transcription guardrail) by sending a later
session.update with turn_detection.create_response: true.
* fix(lint): silence PLR0915 on client_ack_messages
The function exceeded the 50-statement limit (64 > 50) after recent
realtime guardrail additions. Matches the existing project pattern for
inherently complex event/message-mapping methods (see _process_event,
translate_messages_to_responses_input, transform_realtime_response,
_arealtime, etc.).
* fix(gemini realtime): preserve original setup config on follow-up session.update
Gemini Live treats a second BidiGenerateContentSetup as a full session
replacement, not a partial merge. The guardrail-driven turn_detection-only
session.update was emitting a setup containing only model + realtimeInputConfig,
which would silently drop tools, generationConfig, inputAudioTranscription, and
systemInstruction from the original setup. Carry forward the cached original
setup and only override realtimeInputConfig.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(realtime): avoid double-serialization and normalize non-dict turn_detection in guardrail override
- Skip the force-override block when the injection block already ran for
the same session.update to avoid redundant JSON re-serialization.
- Normalize non-dict client-provided turn_detection values (flat and
nested audio.input.turn_detection) to a dict before enforcing
create_response=False, matching the injection block's behavior and
preventing potential bypass on backends that accept non-dict values.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* test(gemini realtime): exercise toolCall → function_call_output name round-trip
Update test_gemini_realtime_function_call_output_transformation to pre-load
the call_id → name mapping by transforming a Gemini toolCall first, then
assert that the resulting Gemini toolResponse functionResponses entry
carries the function name. This pins the production round-trip rather
than the degenerate 'name missing' branch.
* fix(realtime): correct conversation_id, VAD disable, modality state, empty toolCall
- Gemini tool-call response.done now includes conversation_id so clients
can match it against the preceding response.created.
- Vertex AI setup no longer overrides an explicit guardrail-injected
create_response: False back to disabled: False; the guardrail's intent
to disable VAD auto-response is now respected.
- Modality handler is now passed the locally-updated response/item IDs
rather than the original input snapshot, preventing stale IDs after a
prior tool-call/response.done in the same JSON message resets them.
- Skip emitting orphaned response.created/response.done events when
Gemini sends an empty functionCalls array.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(realtime): preserve client session.update fields on follow-up Gemini setup
In non-deferred mode the auto-setup pre-populates session_configuration_request,
so a later client session.update carrying tools or instructions used to fall
into the subsequent path and only forward turn_detection. Rebuild a merged
follow-up setup that overlays the new client fields on top of the original
setup so tools/instructions/etc. are no longer silently dropped.
* fix(gemini realtime): include usage on tool-call response.done; coerce non-dict tool output to struct
- Tool-call response.done now includes an empty usage object, matching the
non-tool-call path so OpenAI-compatible clients always see usage.
- _handle_function_call_output wraps non-dict JSON parses under a 'result'
key so Gemini's functionResponses[].response (a Struct) always receives a
mapping.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): deep-merge nested config in follow-up session update
Previously, the follow-up setup performed a shallow merge between the
original setup and new overrides. If a session.update touched any field
inside generationConfig (e.g. modalities), the entire generationConfig
would be replaced, silently dropping unrelated sub-keys like temperature
or maxOutputTokens. Apply the same deep-merge to realtimeInputConfig so
partial automatic-activity-detection updates don't drop other realtime
input config fields either.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): default conversation_id before tool-call response.done
mypy flagged that response.done's conversation_id (str on the TypedDict)
could be None when current_response_id was already set on entry. Ensure
the fallback runs unconditionally before the response is constructed.
* fix(realtime): deep-merge generationConfig and refresh cache on follow-up setup
A subsequent Gemini session.update that touches any generationConfig sub-field
(e.g. just temperature) was clobbering the original generationConfig — silently
dropping responseModalities and switching the session to text-only. Deep-merge
generationConfig so existing keys (responseModalities, maxOutputTokens, ...) are
preserved when the client updates only a subset.
Also drop the early-return in _cache_session_configuration_request so the
cached payload tracks the latest setup sent to the backend. Without this,
downstream readers (transform_session_created_event, modality lookup in
return_new_content_delta_events) keep reading stale modalities/system
instruction after a follow-up setup.
* fix(gemini realtime): mirror modalities/temperature/max_output_tokens on tool-call response.created
The audio/text response.created preamble includes modalities, temperature,
and max_output_tokens on the response object so spec-compliant clients can
initialise per-response state. The tool-call response.created was missing
these fields, leaving clients without consistent response metadata when a
response starts with a tool call instead of content. Read them from the
cached session_configuration_request the same way the audio/text path
does.
* fix(gemini realtime): keep call_id→name mapping across function_call_output retries
A client SDK that retries function_call_output (or sends the same result
twice) would previously hit a missing-name lookup on the second send
because _handle_function_call_output popped the call_id → name entry.
Without name, Gemini may silently reject the response. Use dict.get so
the mapping persists for the lifetime of the session.
* fix(gemini realtime): empty toolCall must not terminate the WebSocket
If Gemini sends a toolCall whose functionCalls list is empty (or absent),
the previous `continue` left returned_message empty and the
"Unknown message type" guard fired, killing the WebSocket session.
Return a normal (empty) result instead so the session keeps going.
* fix(vertex realtime): warn when dropping guardrail turn-detection update
In non-deferred mode the auto-setup is sent on connect, so the audio-transcription
guardrail's subsequent session.update carrying turn_detection.create_response=False
cannot be forwarded as a second setup (Vertex Live closes the WebSocket with 1007).
Surface a warning when this specific drop happens so operators know the model
will auto-respond before the guardrail can gate it, instead of failing silently
at debug level.
* fix(gemini realtime): deep-merge automaticActivityDetection on follow-up session.update
The follow-up setup merge already deep-merged generationConfig and
realtimeInputConfig, but realtimeInputConfig.automaticActivityDetection
itself is a nested dict. A partial VAD update (e.g. the
guardrail-injected disabled=True from create_response=False) silently
dropped unrelated knobs such as silenceDurationMs and prefixPaddingMs
from the original setup. Deep-merge that block too so partial overrides
only touch the fields they specify.
* fix(realtime): record synthetic session.created in deferred-setup mode
The deferred-setup path emits a synthetic session.created directly to
the client websocket but did not run it through RealTimeStreaming's
store_message, so the event was missing from the session log used by
success_handler / async_success_handler. Call store_message before
forwarding so the synthetic event lands in the same log stream as
provider-driven events.
* fix(gemini realtime): bound _tool_call_id_to_name with an LRU; exercise modality forwarding test
Two minor follow-ups from review:
* Switch _tool_call_id_to_name to a 256-entry LRU OrderedDict so a long
session with many tool calls doesn't grow the dict without bound,
while retried function_call_output lookups still hit for recently-seen
call_ids.
* Fix test_gemini_realtime_transformation_session_created to wrap the
cached session config in {"setup": ...} so the modality lookup in
transform_session_created_event actually exercises responseModalities
forwarding (the prior payload was silently treated as empty).
* test(gemini realtime): wrap remaining cached session configs in setup envelope
The session_configuration_request the proxy caches is always serialized
as {"setup": ...}; three modality-related tests dumped a bare config
dict instead, so transform_session_created_event's
`.get('setup', {})` quietly returned an empty dict and the
responseModalities lookup ran against the default rather than the
fixture. Wrap the remaining tests in the same shape the production
cache uses so any regression in modality forwarding actually trips.
* fix(gemini realtime): cast merged realtimeInputConfig for typeddict assignment
mypy flagged the assignment of the merged dict into
BidiGenerateContentSetup.realtimeInputConfig with [typeddict-item]: the
intermediate variable widens to dict[Any, Any], losing the TypedDict
narrowing the previous dict-literal form had.
* test(gemini realtime): wrap test_gemini_tool_call_resets_ids fixture in setup envelope
The cached session_configuration_request the proxy stores is always
serialized as {"setup": ...}; this test passed a bare config dict, so
transform_session_created_event's .get('setup', {}) returned an empty
dict and the responseModalities lookup ran against the default rather
than the fixture. Wrap the fixture in the same shape the production
cache uses.
* fix(gemini realtime): skip unknown sibling keys in transform loop
Gemini realtime messages can include sibling metadata keys like
usageMetadata alongside primary payload keys (toolCall, serverContent).
Previously, the transform loop called map_openai_event for every
top-level key, raising ValueError for unknown ones and terminating
the WebSocket session.
Skip top-level keys not present in MAP_GEMINI_FIELD_TO_OPENAI_EVENT
to keep the session alive when Gemini emits usage metadata with a
toolCall response.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): scope dotted-key event lookup and propagate session metadata to tool-call response.done
- map_openai_event: only check the current key/value pair when resolving
dotted map entries (e.g. serverContent.turnComplete) so a sibling key in
the same frame can't misclassify the event being processed
(e.g. toolCall returning RESPONSE_DONE).
- tool-call path: extract generationConfig once and include modalities,
temperature, and max_output_tokens on response.done so its shape matches
response.created and the non-tool-call response.done.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): cast maxOutputTokens to int for typeddict assignment
* fix(gemini realtime): use camelCase maxOutputTokens in response.done
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): cast maxOutputTokens to int for typeddict assignment
* fix(realtime): inject guardrail turn_detection on subsequent session.update without one
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): tolerate sibling-only frames (e.g. standalone usageMetadata)
A Gemini Live frame that contains only metadata keys outside
_KNOWN_GEMINI_TOP_LEVEL_KEYS (e.g. a bare {"usageMetadata": {...}}
emitted between turns) leaves returned_message empty after the
transform loop and was tripping the 'Unknown message type' guard,
which raised ValueError and terminated the WebSocket session.
Treat such frames as no-ops and return the unchanged state instead.
* fix(gemini realtime): preserve sibling toolCall when serverContent has only transcription
Previously, when a Gemini frame contained both a transcription-only
serverContent and a sibling toolCall, the transcription handler would
early-return and silently drop the toolCall. Instead, mark serverContent
as handled and fall through so the main loop still processes siblings
like toolCall, while preserving the prior no-op behavior for empty/
transcription-only frames.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* refactor(gemini realtime): drop unused json_message arg from map_openai_event
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): promote nested turn_detection when flat value is not a dict
When the session payload had `turn_detection: None` (or any non-dict value), the
normalizer skipped promoting the GA nested `audio.input.turn_detection` because
it only checked key presence. The stale None then flowed into
`map_automatic_turn_detection` and raised TypeError on `'create_response' in value`.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(realtime): run guardrails on function_call_output content
Tool result outputs are client-controlled and fed to the model, so
they must pass the same content checks as user text messages.
Otherwise an attacker can smuggle blocked content into a
function_call_output and have the model process it.
* fix(gemini realtime): emit function_call_arguments.delta before .done
Gemini delivers the full function-call arguments in a single toolCall
frame. The OpenAI Realtime spec orders the streaming events as
output_item.added -> function_call_arguments.delta(+) ->
function_call_arguments.done -> output_item.done. Emit a single delta
carrying the complete arguments string before the matching .done so
spec-compliant SDK clients that accumulate deltas and gate finalisation
on at least one delta arriving do not stall on Gemini tool calls.
* fix(realtime): avoid stale session.created flag triggering guardrail re-injection
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(ci): restore guardrail injection on duplicate session.created and cast realtime delta event
- Re-enable the one-time guardrail turn_detection update on duplicate
session.created. `_maybe_send_guardrail_turn_detection_update` is
already idempotent via `_guardrail_turn_detection_update_sent`, so
the previous guard was unnecessary and broke the deferred-setup path
where the synthetic session.created is emitted by llm_http_handler
outside this loop (no prior chance to inject).
- Cast the response.function_call_arguments.delta dict appended to
`returned_message: List[OpenAIRealtimeEvents]` so mypy is satisfied.
* fix(realtime): forward sanitized function_call_output on guardrail block
Providers that pair every toolCall with a toolResponse (e.g. Gemini and
Vertex Live) stay in the awaiting-tool-call state until a toolResponse
arrives. Dropping a blocked function_call_output outright left those
providers stalled — the subsequent guardrail clientContent and
response.create were ignored because the prior toolCall had no matching
toolResponse.
When the client-supplied tool output fails the realtime guardrail check,
forward a sanitized placeholder function_call_output (same call_id,
generic policy marker as output) instead of dropping the message
entirely. The placeholder carries no blocked content, so the model never
sees it, while still completing the provider's tool-call cycle so the
session can recover and the violation message reaches the user.
* fix(gemini realtime): preserve sibling keys on empty toolCall no-op
Replace the early return on `functionCalls` empty/absent with a
`continue` plus a `tool_call_handled` flag that mirrors the existing
`server_content_handled` pattern. The post-loop guard already
distinguishes intentionally-consumed known keys from genuinely-unknown
messages, so adding `toolCall` to that exclusion list lets the loop
continue iterating over any sibling top-level keys in the same Gemini
frame instead of short-circuiting on the first empty toolCall.
In practice Gemini's protobuf places `toolCall`/`serverContent`/
`setupComplete` in a `oneof` so the only realistic sibling is
`usageMetadata` (already filtered as unknown-top-level), but the
uniform handling avoids silently discarding any future sibling key
should the wire format grow.
* fix(gemini realtime): redact realtime payloads from debug logs
The transform_realtime_response debug logs were dumping the raw inbound
Gemini frame and each outbound OpenAI event payload (up to 500 chars).
Realtime frames carry transcripts, model output, and tool-call arguments,
so those strings ended up in application logs whenever DEBUG was enabled.
Replace the inbound dump with just the top-level frame keys and the
outbound dump with just the event type.
* fix(realtime): check function_call_output before user role to prevent guardrail bypass
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): propagate usageMetadata on tool-call response.done
Gemini Live emits usageMetadata as a sibling top-level key alongside the
toolCall frame; the tool-call branch was unconditionally building
response.done from get_empty_usage(), so tokens consumed by tool-call
turns were recorded as zero spend and bypassed LiteLLM budget
accounting. Mirror the non-tool-call RESPONSE_DONE path: when the same
frame carries usageMetadata, run VertexGeminiConfig._calculate_usage and
forward the real token counts.
* fix(realtime): send sanitized toolResponse before guardrail clientContent
Two related fixes for the function_call_output blocked-by-guardrail path:
1. Ordering: Gemini Live requires a matching toolResponse immediately
after a toolCall before any other client message. Previously we ran
the guardrail first (which sends clientContent/cancel) and only then
forwarded the sanitized function_call_output. Add an optional
pre_block_backend_message arg to run_realtime_guardrails so the
sanitized toolResponse is emitted before the guardrail's own backend
messages.
2. Stale pending flag: stop setting _pending_guardrail_message in the
tool-output block. That flag exists to swallow the reflexive
response.create an OpenAI client sends right after a user text
message. In tool-calling flows the client may never send a
response.create (e.g. Gemini SDKs auto-respond), so leaving the flag
set would consume an unrelated response.create from a later turn.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* test(model_prices): allow audio_transcription_config in schema
* fix(gemini realtime): event_id, item copy, and dict guard for tool-call events
- Emit event_id on response.output_item.added for tool calls so spec-compliant
OpenAI Realtime SDK clients can index/deduplicate the event like every other
server-sent event in the sequence.
- Pass a shallow copy of function_call_item to response.output_item.done and
conversation.item.created so downstream handlers (e.g. the beta-protocol
translator) that mutate the item dict don't corrupt sibling events sharing
the same reference.
- Guard map_openai_event against non-dict values (e.g. Gemini's
'setupComplete: true' boolean payload) so the WebSocket session doesn't die
with an AttributeError on the unguarded .get() call.
Add NotRequired event_id field on OpenAIRealtimeStreamResponseOutputItemAdded
to keep existing call-sites that don't set event_id compatible.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(gemini realtime): buffer standalone usageMetadata for next response.done
Gemini Live can emit usageMetadata as a standalone WebSocket frame between
turns. The previous transformer treated those frames as no-ops, so token
counts arriving outside the closing turnComplete/toolCall frame were
dropped from spend and budget accounting. An authenticated client could
drive turns whose usage was recorded as zero, bypassing budgets.
Buffer any standalone usageMetadata on the config instance and attribute
the deferred counts to the next emitted response.done (tool-call or
normal). In-frame usageMetadata remains authoritative and clears the
buffer.
* merge main (#28839)
* fix(helm): drop main- prefix from default image tag (#28710)
* fix(helm): drop main- prefix from default image tag
The default image tag in the deployment + migrations-job templates was
`main-{{ .Chart.AppVersion }}`. The current release pipeline publishes
content tags without the `main-` prefix (e.g. `v1.85.1` / `1.85.1`,
`v1.86.0-rc.1` / `1.86.0-rc.1`), so the rendered ref points at a tag
that does not exist on GHCR or DockerHub and installs fail with
ImagePullBackOff.
- templates/deployment.yaml, templates/migrations-job.yaml: render
`.Chart.AppVersion` directly instead of `main-<AppVersion>`.
- Chart.yaml: bump stale `appVersion: v1.80.12` (not on either
registry) to `v1.85.1` so local-checkout installs also resolve.
- values.yaml: update the commented tag-override hint to match.
* fix(helm): use :latest in tag override example, not pinned version
Per review: ghcr.io/berriai/litellm-database:latest is a floating
alias for the most recent stable (same digest as :main-stable),
maintained by the release pipeline's UPDATE_LATEST advance step.
Better example than a pinned version that goes stale.
* test(model_prices): allow audio_transcription_config in schema (#28708)
The schema in test_aaamodel_prices_and_context_window_json_is_valid uses
additionalProperties: false. The azure/speech/azure-stt entry added in
#27482 introduced an audio_transcription_config field that the schema
did not whitelist, so the test fails on every branch built on top of
staging.
Add the field as a string property.
* fix(team): refresh team cache on team_model_add/delete (LIT-3244) (#28683)
* fix(team): refresh team cache on team_model_add/delete (LIT-3244)
team_model_add and team_model_delete wrote to the DB but did not
invalidate the in-memory LiteLLM_TeamTableCachedObj used by
common_checks. After the v1.83.14 common_checks centralization made
team.models authoritative on /v1/files and /v1/vector_stores/*,
adding a Team-BYOK model silently failed to grant the new public
model name to team members until the cache TTL expired (and a
removed model kept working until then on the symmetric path).
Extract the cache-refresh snippet from update_team into a small
helper and apply it consistently at all three team-write sites.
* test: also assert updated models in team-cache-refresh pin
Strengthens the LIT-3244 regression test to also assert
`call_kwargs["team_table"].models` matches the updated row,
not just `team_id`. Both `existing_team` and `updated_team`
share `team_id` in the test setup, so the previous assertion
would have passed even if the implementation accidentally cached
the pre-mutation row.
Greptile review feedback.
* fix(team): hydrate object_permission on cache-refreshing team updates
The Prisma update calls in update_team, team_model_add, and
team_model_delete returned a team row with object_permission_id set
but object_permission=None (the relation was not requested via
include=). _refresh_cached_team then wrote that to the in-memory
LiteLLM_TeamTableCachedObj, and the cache-hit path in get_team_object
returns the cached object without re-hydrating. Downstream consumers
(validate_key_search_tools_against_team, the MCP/agent authz paths)
treat a missing object_permission as no team-level restriction, so
a team-write op silently dropped object-permission enforcement until
the cache TTL expired or a DB-fetch path re-hydrated it.
Add include={"object_permission": True} to all three updates so the
refresh writes a complete cached team. Extend the LIT-3244 regression
test to pin both the cached object_permission and the include shape
on the Prisma call.
Surfaced in PR review of LIT-3244.
* fix(ui/add-model): stop vertex_ai-anthropic_models from leaking under Anthropic (#28723)
`getProviderModels()` matched a model into a provider's dropdown when the
model's `litellm_provider` string *contained* the provider key as a
substring. The intent was to admit suffix variants (e.g. `anthropic_text`,
`bedrock_converse`), but the substring check is too loose: it also pulls in
unrelated providers whose name happens to contain the key, most visibly
`vertex_ai-anthropic_models` matching `anthropic` and `vertex_ai-openai_models`
matching `openai`.
Replace `.includes()` with separator-anchored prefix matching
(`startsWith(provider + "_")` / `startsWith(provider + "-")`). All legitimate
variants in `model_prices_and_context_window.json` still match
(`anthropic_text`, `azure_text`, `azure_ai`, `bedrock_converse`,
`bedrock_mantle`, `cohere_chat`, `fireworks_ai-embedding-models`,
`vertex_ai-*`, `vertex_ai_beta`), and the cross-provider leak is closed.
Tests: update one assertion that pinned the buggy substring behavior
(`custom_openai_endpoint` matching `openai` — not a real provider value);
add 6 new tests covering the leak regressions and the variant-preservation
contract for vertex_ai/bedrock/fireworks.
* Fix spend logs v2 route permissions (#28705)
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: ryan-crabbe-berri <ryan-crabbe-berri@users.noreply.github.com>
* fix(proxy): Bedrock Knowledge Base pass-through: preserve SigV4 headers and signed request body (#27526)
* Fix Bedrock KB pass-through SigV4 headers and signed body
Coerce botocore HeadersDict to a dict for pass-through routes. When
forward_headers is true, drop request headers that collide case-insensitively
with signed headers so client Bearer auth does not shadow AWS SigV4.
Send prepped.body as raw content so the outbound payload matches the
signature after logging hooks mutate the parsed dict.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Simplify pass-through raw body handling
Read the SigV4-signed bytes directly from request.state inside
pass_through_request instead of threading a custom_raw_body argument
through three functions. Helper methods are restored to their original
signatures, and the new branch lives in one place at each httpx call site.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Harden pass-through raw body read from request.state
Guard missing request.state (test fixtures) and ignore non-bytes/str
values so MagicMock does not trigger the SigV4 raw-body path.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Test pass_through_request state_raw_body uses httpx content=
Cover non-streaming (async_client.request) and streaming (build_request)
paths so SigV4 bytes on request.state are not replaced by json= of a
hook-mutated dict.
Co-authored-by: Cursor <cursoragent@cursor.com>
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
* chore(tests): migrate Bedrock CI to AWS account 941277531214 (#28728)
* chore(tests): migrate Bedrock CI from AWS account 888602223428 to 941277531214
The original account (888602223428) was put under a security restriction by
AWS after a root access key leaked in a PR comment. While that account works
its way through the AWS Support unlock process, Bedrock-touching CI tests have
been migrated to a fresh account (941277531214).
Changes:
- Replace 26 hardcoded references to 888602223428 with 941277531214 across
8 files (provisioned-model ARNs, imported-model ARNs, AgentCore runtime
ARNs, batch execution role ARN, and example proxy config).
- The provisioned-model and imported-model ARNs are referenced only from
mocked unit tests — no AWS resources to recreate.
- The batch execution IAM role has been recreated in the new account with
the same name and equivalent permissions.
- The two AgentCore runtimes (hosted_agent_r9jvp-3ySZuRHjLC,
hosted_agent_13sf6-cALnp38iZD) are being recreated in the new account
under the same names — see tools/agentcore-deploy/ in a follow-up.
CircleCI env vars AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY / AWS_REGION_NAME
were updated separately via the CircleCI API to point at the new account.
Smoke-tested locally against the new account:
aws bedrock-runtime converse --region us-west-2 \
--model-id us.anthropic.claude-sonnet-4-5-20250929-v1:0 \
--messages '[{"role":"user","content":[{"text":"ping"}]}]'
→ 200, model returned 'pong'
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* chore(tests): refresh AgentCore ARN suffixes to match newly-deployed runtimes
The first migration commit replaced just the account ID, but AgentCore
auto-assigns a random 10-char suffix to every runtime on creation — we
can't reuse the original suffixes (`3ySZuRHjLC`, `cALnp38iZD`) in the
new account. Updated the AgentCore-runtime ARNs in the three files that
reference real runtime IDs (not the mock-based unit-test ARNs).
Deployed runtimes:
arn:aws:bedrock-agentcore:us-west-2:941277531214:runtime/hosted_agent_r9jvp-Rq79QFC2fp
arn:aws:bedrock-agentcore:us-west-2:941277531214:runtime/hosted_agent_13sf6-4046UzHSwy
Both runtimes are status=READY and pass a smoke invoke:
$ aws bedrock-agentcore invoke-agent-runtime --agent-runtime-arn ... --payload '{"prompt":"ping"}'
→ 200, {"result": "echo: ping"}
The agent is a minimal echo (see /tmp/agentcore_deploy/agent.py for the
deploy artifacts). Tests that only verify the SDK wiring will pass; if any
test asserts on agent output content, swap the echo for the real agent.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* chore(tests): point Bedrock batch tests at new-account S3 bucket
The account migration (888602223428 -> 941277531214) was a flat
account-ID swap, which only rewrites ARNs that embed the account
number. S3 bucket names carry no account ID, so the live Bedrock
batch tests still uploaded to `litellm-proxy` — a bucket that lives
in the old account. S3 names are globally unique, and the old account
still holds that name, so it can't be recreated in the new account.
Rename to `litellm-proxy-941277531214` (account-ID suffix guarantees
global uniqueness). The bucket must be created in 941277531214 and the
batch execution role granted s3:GetObject/PutObject/ListBucket on it
before this job is run in CI.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore(tests): point live S3 logging test at new-account bucket
Same account-ID-free blind spot as the batch bucket: `load-testing-oct`
lives in the old account and its name can't be reused globally. The
`logging_testing` CI job is wired into the workflow and runs
test_basic_s3_logging, which uploads to this bucket with the CI env
creds, then lists and deletes objects — a live dependency.
Rename to `load-testing-oct-941277531214`. The bucket must exist in the
new account with the CI IAM principal granted
s3:PutObject/GetObject/ListBucket/DeleteObject before this job runs.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore(tests): repoint Bedrock guardrail IDs to new-account guardrails
The migration left guardrail IDs untouched (no account ID in them), so
all live guardrail tests failed with "guardrail identifier or version
does not exist" against 941277531214. Recreated both guardrails in the
new account and updated the hardcoded IDs:
- wf0hkdb5x07f -> zgkmukebruil (PII mask: PHONE + CREDIT_DEBIT_CARD,
with explicit inputAction=ANONYMIZE so masking applies to INPUT,
which is the source litellm's moderation hook sends)
- ff6ujrregl1q -> 4w3d1di3snt5 (blocks "coffee"; blocked message set
to the exact string the tests assert on)
Updated test_bedrock_guardrails.py, otel_test_config.yaml, and the
guardrailConfig in test_bedrock_completion.py. Verified locally: the 5
previously-failing guardrail tests now pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(bedrock): migrate legacy models to current inference profiles
The new CI account (941277531214) cannot invoke legacy Bedrock models
(AWS gates them: "marked by provider as Legacy... not actively using in
the last 30 days"). Migrated the live-call tests:
- anthropic.claude-3-sonnet-20240229 -> us.anthropic.claude-sonnet-4-5-20250929-v1:0
- anthropic.claude-3-haiku-20240307 -> us.anthropic.claude-haiku-4-5-20251001-v1:0
Current Claude models on Bedrock require the us. inference-profile prefix
(bare on-demand ids are rejected).
cohere.command-r-plus has no working replacement (all Cohere is legacy-
gated in the new account): swapped to claude-haiku-4-5 in provider-
agnostic param lists. amazon.titan-image-generator skipped (no working
replacement). Mocked/transformation/cost tests that reference the legacy
strings are intentionally left unchanged. Verified live against the new
account.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(bedrock): repoint SageMaker + Knowledge Base to new-account resources
These referenced account-scoped resources by hardcoded id that only
existed in the old account, so the migration's account-ID swap missed
them. Recreated in 941277531214 and repointed:
- SageMaker endpoint jumpstart-dft-hf-textgeneration1-mp-20240815-185614
-> litellm-ci-textgen (gpt2 on a TGI container, ml.g5.xlarge)
- Bedrock Knowledge Base T37J8R4WTM -> LCYXFBR2TU (OpenSearch Serverless
vector store + titan-embed-text-v2, seeded with a LiteLLM doc)
Verified live: test_sagemaker.py (12 passed) and
test_bedrock_knowledgebase_hook.py (12 passed).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(reasoning_effort_grid): skip bedrock claude-opus-4-7 cells (not entitled on 941277531214)
claude-opus-4-7 is listed in the new Bedrock CI account's foundation
models but invoke is denied (AccessDeniedException: "not available for
this account"). Bedrock access to the flagship Opus requires an AWS
Sales request, not the self-serve model-access toggle, so it can't be
enabled inline with the rest of the account migration.
Add an optional `skip_reason` to ModelEntry and set it on the
bedrock-claude-opus-4-7 entry; the grid test honors it via pytest.skip.
Cell count (231) and route coverage are unchanged, so the structural
asserts still pass. Restore coverage by deleting the one skip_reason
line once access is granted.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(bedrock): swap/skip legacy-gated models unavailable on new CI account
The migrated AWS account (941277531214) cannot access several models that
the old account could, so the remaining red CI jobs were hitting real
Bedrock "Access denied / Legacy" and "account not authorized" errors:
- image_gen: skip both Nova Canvas test classes (amazon.nova-canvas-v1:0 is
legacy-gated), matching the existing titan skip.
- batches: skip test_async_file_and_batch (Bedrock batch inference is not
authorized on the new account; requires an AWS support case).
- litellm_overhead: swap legacy claude-3-5-haiku for the active
us.anthropic.claude-haiku-4-5 inference profile.
- test_completion_claude_3_function_call: swap legacy claude-3-sonnet for the
active us.anthropic.claude-sonnet-4-5 inference profile.
https://claude.ai/code/session_01Y7zgHYu9GX29YRwV4yiWAa
* test(bedrock): fix remaining e2e legacy-model + batch failures on new CI account
- e2e_openai_endpoints: skip test_bedrock_batches_api (Bedrock batch inference
is not authorized on account 941277531214) and migrate the missed
s3_bucket_name in oai_misc_config.yaml to litellm-proxy-941277531214.
- build_and_test: swap legacy bedrock claude-3-sonnet for the active
us.anthropic.claude-sonnet-4-5 inference profile in the proxy structured
output e2e test.
https://claude.ai/code/session_01Y7zgHYu9GX29YRwV4yiWAa
* test(bedrock): make opus-4-7 + batch cells fail loudly and mock image-gen (#28791)
Replace the silent skips added for the new CI account with noisier behavior:
- reasoning-effort grid: opus-4-7 cells now fail (when AWS creds are present)
instead of skipping, so the missing entitlement stays visible in CI; they
still skip when AWS creds are absent (local dev)
- Bedrock batch inference tests: drop the skip so they run and fail until
batch access is granted
- Titan + Nova Canvas image-gen tests: mock the Bedrock HTTP call so the
transform + cost-tracking path stays under test without live model access
https://claude.ai/code/session_01MT7SWDnXUjv6e6EPG7BDjT
Co-authored-by: Claude <noreply@anthropic.com>
* test(bedrock): use pytest.xfail for known-failing opus-4-7 cells
Replace pytest.fail with pytest.xfail when a model has a fail_reason,
so known-broken cells stay visible as XFAIL without keeping CI red.
Co-authored-by: Yassin Kortam <yassin@berri.ai>
---------
Co-authored-by: Mateo <mateo@Mateos-MacBook-Pro.local>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(otel): export SERVER span on management-endpoint success without http_request (#28794)
Co-authored-by: Yassin Kortam <yassinkortam@Yassins-MacBook-Pro.local>
* chore(ci): merge dev branch (#28801)
* chore(proxy): route path-dependent call sites through get_request_route
Replace direct ``request.url.path`` reads in auth, ACL, routing, and
audit-log decisions with ``get_request_route(request)`` — the helper
already added in ``auth/auth_utils.py`` that returns the ASGI
``scope["path"]`` with ``root_path`` stripped. Starlette reconstructs
``url.path`` from the Host header; ``scope["path"]`` is uvicorn's
parse of the request line and matches what FastAPI dispatches on, so
it's the authoritative route for any decision that should agree with
the actual handler.
Sites:
- _experimental/mcp_server/auth/user_api_key_auth_mcp.py
- management_endpoints/mcp_management_endpoints.py
- vector_store_endpoints/utils.py
- pass_through_endpoints/pass_through_endpoints.py
- auth/route_checks.py
- litellm_pre_call_utils.py
- spend_tracking/spend_management_endpoints.py
- common_utils/http_parsing_utils.py
- management_helpers/utils.py
- health_endpoints/_health_endpoints.py
Adds regression tests in tests/proxy_unit_tests/test_proxy_routes.py
that construct a Request with scope["path"] set to a benign route and
the Host header crafted so url.path would resolve differently; each
site's decision is asserted against scope["path"].
* chore(proxy): make get_request_route imports lazy at call sites
Move the ``from litellm.proxy.auth.auth_utils import get_request_route``
imports added in the prior commit back to the function bodies that use
them. The module-level form participates in a long-standing import
cycle through ``auth_utils -> _types -> ...`` and was flagged by CodeQL
on the PR; the lazy form matches the pattern the proxy already uses
for ``user_api_key_auth`` and related helpers elsewhere in these files.
Also drop the ``RouteChecks._is_assistants_api_request`` delegation in
``_get_metadata_variable_name`` introduced in the prior commit — the
delegation pulled ``RouteChecks`` into the same cycle, and the call
site reuses the resolved route for its other branches, so inlining
the substring check is both cycle-free and avoids a redundant second
``get_request_route`` call.
Comment in test_proxy_routes.py acknowledges that the two MCP table
entries exercise ``get_request_route`` directly rather than the full
production handler (which needs ASGI scope + MCP state to invoke).
---------
Co-authored-by: shin-berri <shin-laptop@berri.ai>
Co-authored-by: user <70670632+stuxf@users.noreply.github.com>
* chore(ci): merge dev branch (#28657)
* feat(dashboard): navbar hierarchy + Agent Platform notifications (#27543)
* feat(dashboard): refine navbar zones and Agent Platform notice
Restructure the admin navbar for production users: clear product vs community
vs personal columns with vertical dividers, icon-only Slack/GitHub in a
shared chip, and Docs/Blog typography aligned on an 8px rhythm.
Add a notifications bell with popover linking to the LiteLLM Agent Platform
repo and optional mark-as-read persistence.
Promote the account control with initials avatar, single-line display name,
and navDisplayName mapping for placeholder user ids (e.g. default_user_id).
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(dashboard): address PR review — AntD buttons, public page guard, dedupe regex
- Replace raw <button> with AntD Button in BlogDropdown, NotificationsBell, UserDropdown, and test mock
- Guard NotificationsBell + container behind !isPublicPage to avoid rendering on public pages
- Remove redundant equality checks in navDisplayName (regex already covers them)
- Remove unused `lower` variable after simplification
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: yuneng-jiang <yuneng@berri.ai>
* fix(dashboard): drop dead useHealthReadiness import in navbar
The module was removed in #27896 (replaced by useHealthReadinessDetails),
but the import survived the rebase. The symbol is unused — only
useHealthReadinessDetails is consumed in the file. Removing the dead
import unblocks the UI TypeScript build.
* fix(dashboard): align CommunityEngagementButtons test with icon-only aria-labels
The component was refactored to an icon-only chip with aria-label='LiteLLM
on GitHub' (squash #27543), but the test still asserted /star us on
github/i. Update the query to match the rendered accessible name.
* refactor(dashboard): drop unused props from NavbarProps
The navbar refactor moved user identity + dark-mode state to internal
hooks (useAuthorized, useWorker), but the NavbarProps interface still
declared userID, userEmail, userRole, premiumUser, isDarkMode, and
toggleDarkMode as required, forcing every caller to thread them through.
Drop them from the interface and all four call sites (page.tsx,
(dashboard)/layout.tsx, public_model_hub.tsx, navbar.test.tsx). Also
shrinks the destructure in layout.tsx so the now-unused locals stop
being pulled out of useAuthorized().
* refactor(dashboard): use useSyncExternalStore for NotificationsBell dismiss flag
Reads/writes of the litellmHideAgentPlatformBanner key were done
directly inside NotificationsBell via a useEffect + useState pair.
Every other localStorage-backed flag in the dashboard (Disable
ShowPrompts, DisableBouncingIcon, DisableShowNewBadge,
DisableUsageIndicator, DisableBlogPosts) is wrapped in a
useSyncExternalStore hook over localStorageUtils so all mounted
components stay in sync.
Extract useHideAgentPlatformBanner to follow the same shape, swap
NotificationsBell to consume it, and add a regression test that
two sibling bells stay in sync without a remount when one is
dismissed.
* refactor: mask credential fields in proxy settings GET responses (#28682)
* refactor: mask credential fields in proxy settings GET responses
Brings SSO settings, cache settings, and the email/Slack alerting view in
/get/config/callbacks in line with the HashiCorp Vault config-override
pattern, so persisted credentials are not transported back to the UI in
plaintext.
* refactor: harden short-value masking and hoist alerting var constant
Closes two review observations:
- mask_sensitive_keys now replaces short values (below the visible
prefix+suffix length) with an all-mask string instead of returning them
unchanged, so a 1-7 character credential is no longer round-tripped
verbatim.
- _ALERTING_SENSITIVE_VARS is moved out of get_config() to a module-level
constant, matching the analogous _SSO_SENSITIVE_FIELDS and
_CACHE_SENSITIVE_FIELDS in the SSO and cache endpoint files.
---------
Co-authored-by: Krrish Dholakia <krrish+github@berri.ai>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(ui): show 2-decimal precision for max_budget on key overview (#28809)
The Key Info Overview tab's Spend card truncated sub-dollar budgets to
"$0" because formatNumberWithCommas defaults to 0 decimals. The Settings
tab passes 2; align the overview so a $0.10 budget renders as "$0.10".
Resolves LIT-2845
* feat(proxy): allow `llm_api_routes` virtual keys to list MCP servers (#28442)
* feat(proxy): allow llm_api_routes virtual keys to list MCP servers
Add a new `mcp_discovery_routes` group (GET /v1/mcp/server and GET
/v1/mcp/server/{server_id}) and include it in `llm_api_routes` so that
virtual keys configured with `allowed_routes=["llm_api_routes"]` can
discover the MCP servers they have access to. Previously these calls
failed with 'Virtual key is not allowed to call this route. Only allowed
to call routes: [llm_api_routes]'.
The GET handlers already sanitize the response for restricted virtual
keys via `_sanitize_mcp_server_list_for_virtual_key`, stripping
credential-bearing fields (url, headers, env). Write methods
(POST/PUT/DELETE) on the same paths remain gated by the existing
handler-level admin role checks.
The new discovery list is intentionally kept OUT of
`mcp_inference_routes`, so `is_llm_api_route()` still returns False
for these paths — this preserves the existing contract that
DISABLE_LLM_API_ENDPOINTS must not block the Admin UI from listing MCP
servers.
Co-authored-by: ryan-crabbe-berri <ryan-crabbe-berri@users.noreply.github.com>
* refactor(proxy): make MCP discovery carve-out method-aware
Replace the `mcp_discovery_routes` group in `llm_api_routes` with a
method-aware special case inside `is_virtual_key_allowed_to_call_route`.
Virtual keys with allowed_routes=["llm_api_routes"] are now permitted
to call only GET /v1/mcp/server and GET /v1/mcp/server/{server_id} —
non-GET methods and multi-segment admin sub-paths fall through to the
existing 403. This keeps the general llm_api_routes list free of
management paths and avoids accidentally exposing POST/PUT/DELETE
writes through the route-check layer.
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: ryan-crabbe-berri <ryan-crabbe-berri@users.noreply.github.com>
* chore(ci): merge dev branch (#28807)
* chore(proxy): route path-dependent call sites through get_request_route
Replace direct ``request.url.path`` reads in auth, ACL, routing, and
audit-log decisions with ``get_request_route(request)`` — the helper
already added in ``auth/auth_utils.py`` that returns the ASGI
``scope["path"]`` with ``root_path`` stripped. Starlette reconstructs
``url.path`` from the Host header; ``scope["path"]`` is uvicorn's
parse of the request line and matches what FastAPI dispatches on, so
it's the authoritative route for any decision that should agree with
the actual handler.
Sites:
- _experimental/mcp_server/auth/user_api_key_auth_mcp.py
- management_endpoints/mcp_management_endpoints.py
- vector_store_endpoints/utils.py
- pass_through_endpoints/pass_through_endpoints.py
- auth/route_checks.py
- litellm_pre_call_utils.py
- spend_tracking/spend_management_endpoints.py
- common_utils/http_parsing_utils.py
- management_helpers/utils.py
- health_endpoints/_health_endpoints.py
Adds regression tests in tests/proxy_unit_tests/test_proxy_routes.py
that construct a Request with scope["path"] set to a benign route and
the Host header crafted so url.path would resolve differently; each
site's decision is asserted against scope["path"].
* chore(proxy): make get_request_route imports lazy at call sites
Move the ``from litellm.proxy.auth.auth_utils import get_request_route``
imports added in the prior commit back to the function bodies that use
them. The module-level form participates in a long-standing import
cycle through ``auth_utils -> _types -> ...`` and was flagged by CodeQL
on the PR; the lazy form matches the pattern the proxy already uses
for ``user_api_key_auth`` and related helpers elsewhere in these files.
Also drop the ``RouteChecks._is_assistants_api_request`` delegation in
``_get_metadata_variable_name`` introduced in the prior commit — the
delegation pulled ``RouteChecks`` into the same cycle, and the call
site reuses the resolved route for its other branches, so inlining
the substring check is both cycle-free and avoids a redundant second
``get_request_route`` call.
Comment in test_proxy_routes.py acknowledges that the two MCP table
entries exercise ``get_request_route`` directly rather than the full
production handler (which needs ASGI scope + MCP state to invoke).
---------
Co-authored-by: shin-berri <shin-laptop@berri.ai>
Co-authored-by: user <70670632+stuxf@users.noreply.github.com>
* fix(team): keep team_alias cache in sync on _cache_team_object writes (#28737)
* fix(team): keep team_alias cache in sync on _cache_team_object writes
_cache_team_object wrote only to the team_id:<id> cache key, but the
JWT auth path that uses team_alias_jwt_field reads from a separate
team_alias:<alias> key (get_team_object_by_alias caches under both
keys on miss, but reads only the alias-keyed one). After any
team-mutation endpoint (team_model_add, team_model_delete,
update_team, the two access-group writes) the team_id cache was
refreshed but the team_alias cache stayed stale until TTL — JWT
callers using team_alias_jwt_field kept seeing the pre-mutation
team for the full cache window.
Mirror the write under the alias key inside _cache_team_object so
every existing caller stays in sync without further changes. Skip
the alias write when team_alias is None/empty so we don't collide
across alias-less teams.
Surfaced testing the LIT-3244 cherry-pick on patch/1.86.0: the
LIT-3244 fix correctly invalidated the team_id cache but the
customer's JWT used team_alias_jwt_field, so they kept hitting the
stale alias-keyed entry.
* fix(team): delete (not overwrite) team_alias cache on _cache_team_object
The prior shape of this PR wrote both team_id:<id> AND team_alias:<alias>
from _cache_team_object. team_alias is NOT unique in the schema
(no @unique on LiteLLM_TeamTable.team_alias), and get_team_object_by_alias
enforces uniqueness on its own DB-fetch path (len(teams) > 1 raises).
Writing the alias-keyed cache from the generic refresh path bypassed
that check: a team admin renaming their team to collide with another
team's alias could silently overwrite the cached team for JWT-by-alias
auth, swapping the resolved team under that alias for the cache window.
Switch the alias-keyed operation from a write to a delete (mirroring
the dual-cache delete pattern in _delete_cache_key_object). After every
team write, the next JWT-by-alias reader cache-misses and falls through
to get_team_object_by_alias, which (a) re-fetches the fresh team from
DB, closing the LIT-3244 staleness gap that motivated this PR, and
(b) enforces alias uniqueness before populating either cache key.
team_id:<id> writes are unchanged — team_id is the table PK and is
guaranteed unique.
Surfaced in veria-ai review on #28739.
* fix(managed-files): anchor model_id regex so it doesn't match llm_output_file_model_id
extract_model_id_from_unified_id used `re.search(r"model_id,([^;]+)", ...)`
which substring-matches the `model_id,` inside the file-ID encoding's
`llm_output_file_model_id,<deployment_uuid>` field. parse_unified_id
then fed that deployment UUID back into the auth path as a model
candidate via _extract_models_from_managed_resource_id, and every
team-BYOK file attach 403'd with:
team not allowed to access model. This team can only access
models=['openai/*']. Tried to access <deployment-uuid>
The team's models list correctly contains the public name (`openai/*`)
that target_model_names matches, but the bogus UUID candidate fails
the wildcard check first.
Anchor the regex to a field boundary (`(?:^|;)model_id,`) so it
matches the legitimate top-level `model_id,<value>` field on
vector_store unified IDs and skips substring matches inside other
fields. File-IDs (which have no top-level `model_id` field) now
return None and contribute no spurious UUID candidate.
Surfaced reproducing LIT-3244 on patch/1.86.0 with the customer's
exact flow: team with openai/* BYOK deployment, JWT-scoped user,
POST /v1/vector_stores/{id}/files attaching a file uploaded with
target_model_names=openai/gpt-4o.
* fix(proxy): hydrate wildcard discovery credentials (#28284) (#28822)
* fix(proxy): hydrate wildcard discovery credentials
* fix(proxy): constrain wildcard credential hydration
Co-authored-by: Dibyo Mukherjee <dibyo@adobe.com>
* ci: add daily oss-agent-shin branch creation workflow (#28829)
Creates litellm_oss_agent_shin_MM_DD_YYYY from main every day at 00:00 UTC.
Lets us retarget oss-agent-shin fork PRs onto a canonical branch so CircleCI runs with secrets, without granting the agent write access.
Co-authored-by: shin-berri <shin-laptop@berri.ai>
Co-authored-by: yuneng-jiang <yuneng@berri.ai>
Co-authored-by: Ishaan Jaffer <ishaanjaffer0324@gmail.com>
* test(proxy): add harness for proxy_server.py behavior-pinning (#28827)
* test(proxy): add harness for proxy_server.py behavior-pinning
Creates tests/test_litellm/proxy/proxy_server/ with:
- conftest.py: 11 shared fixtures (app, client, mock_prisma, auth_as,
mock_router with parametrized response builders, normalize, etc.)
- _coverage_check.py: per-PR coverage gate (line + branch) against a
baseline, self-selects target by inspecting which placeholder files
have been filled
- _pin_check.py: AST-based gate that verifies every pin-list item has
>=1 happy + >=1 error test with a real assertion (no status-only)
- test_harness_smoke.py: 19 smoke tests covering every fixture +
both scripts end-to-end
- 26 placeholder test files (one docstring each) reserved for
follow-up PRs per the directory ownership in the Notion plan
- .coverage_baseline pinned at 0% so future PRs measure deltas
against new-tests-only and aren't entangled with the broader
scattered test suite
Adds a dedicated proxy-server job to test-unit-proxy-endpoints.yml
so this directory's runtime + coverage are tracked independently.
Plan: https://www.notion.so/36c43b8acdab81ee845fd5365128a2fc
* ci(proxy-endpoints): allow workflow_dispatch
Lets the workflow be triggered manually on a branch via
`gh workflow run`, which is needed for the verify-first
flow on workflow changes before opening a PR.
* test(proxy): address review feedback on proxy_server harness
- conftest.py: anchor sys.path insert to __file__ (Path(__file__).resolve().parents[4])
instead of CWD-relative os.path.abspath("../../../../") which resolved
to the wrong directory when pytest is launched from the repo root.
- _coverage_check.py: actually read .coverage_baseline and use it as
the floor (line_min = max(target, baseline)). Closes the gap between
the PR description's "delta semantics" and what the script was doing.
With baseline=0.0 today this is a no-op; future PRs that update the
baseline cause regressions (test deletions etc.) to trip the gate
even if the static PR target is still met.
- _pin_check.py: drop unreachable startswith("_") guard
(test_*.py glob never yields underscore-prefixed names) and read
each test file once instead of twice.
* feat(openai): apply regional-processing cost uplift for EU/US data residency (#28626)
* feat(openai): apply regional-processing cost uplift for EU/US data residency
OpenAI charges a 10% uplift on the latest GPT models when requests are
served from a regionalized hostname (eu./us.api.openai.com). Infer the
region from `api_base`, expose it on `kwargs["litellm_params"]["data_residency"]`,
and multiply the computed cost by a per-model
`regional_processing_uplift_multiplier_<region>` field.
https://claude.ai/code/session_012ebH44s7ohYxjoix5CXzTW
* test: allow regional_processing_uplift_multiplier_{eu,us} in model_prices schema
* fix(cost): tighten data_residency inference and restore model_cost in tests
- Only infer OpenAI data_residency when custom_llm_provider == "openai";
drop the implicit None fallback so non-OpenAI callers can't accidentally
pick up a regional tag from a stray OpenAI hostname.
- _local_model_cost_map fixture now snapshots and restores
litellm.model_cost and LITELLM_LOCAL_MODEL_COST_MAP so tests don't leak
state across the session.
* refactor(openai): move data_residency helper under llms/openai
* fix: thread data_residency through realtime stream cost calculation
Co-authored-by: Yassin Kortam <yassin@berri.ai>
* fix(cost): thread data_residency through batch_cost_calculator
Apply the OpenAI regional-processing uplift multiplier to retrieve_batch
cost paths so Batch API requests served via eu./us.api.openai.com are
priced at the same uplifted token rates as completions/transcriptions.
* refactor(openai): encapsulate provider check inside infer_openai_data_residency
Move the custom_llm_provider == "openai" guard from get_litellm_params
into the helper itself so the core utility no longer carries
provider-specific dispatch logic. Callers pass through the provider
unconditionally; the helper returns None for any non-OpenAI provider.
* fix(responses): thread data_residency through Responses logging params
The Responses API paths build their logging litellm_params dict after
provider resolution but did not include data_residency, so cost calc
saw None even when the effective api_base was a regional OpenAI host.
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
---------
Co-authored-by: yuneng-jiang <yuneng@berri.ai>
Co-authored-by: ryan-crabbe-berri <ryan@berri.ai>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: ryan-crabbe-berri <ryan-crabbe-berri@users.noreply.github.com>
Co-authored-by: milan-berri <milan@berri.ai>
Co-authored-by: Mateo Wang <277851410+mateo-berri@users.noreply.github.com>
Co-authored-by: Mateo <mateo@Mateos-MacBook-Pro.local>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
Co-authored-by: Yassin Kortam <yassinkortam@Yassins-MacBook-Pro.local>
Co-authored-by: shin-berri <shin-laptop@berri.ai>
Co-authored-by: user <70670632+stuxf@users.noreply.github.com>
Co-authored-by: Krrish Dholakia <krrish+github@berri.ai>
Co-authored-by: Dibyo Mukherjee <dibyo@adobe.com>
Co-authored-by: ishaan-berri <155045088+ishaan-berri@users.noreply.github.com>
Co-authored-by: Ishaan Jaffer <ishaanjaffer0324@gmail.com>
* Revert "merge main (#28839)"
This reverts commit fa956e8c42ea975a452665fedd5c4ecebaab9be8.
* fix(gemini-realtime): prevent double-attribution of usageMetadata in multi-key frames
Co-authored-by: Yassin Kortam <yassin@berri.ai>
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Co-authored-by: yuneng-jiang <yuneng@berri.ai>
Co-authored-by: ryan-crabbe-berri <ryan@berri.ai>
Co-authored-by: ryan-crabbe-berri <ryan-crabbe-berri@users.noreply.github.com>
Co-authored-by: milan-berri <milan@berri.ai>
Co-authored-by: Mateo <mateo@Mateos-MacBook-Pro.local>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Yassin Kortam <yassinkortam@Yassins-MacBook-Pro.local>
Co-authored-by: shin-berri <shin-laptop@berri.ai>
Co-authored-by: user <70670632+stuxf@users.noreply.github.com>
Co-authored-by: Krrish Dholakia <krrish+github@berri.ai>
Co-authored-by: Dibyo Mukherjee <dibyo@adobe.com>
Co-authored-by: ishaan-berri <155045088+ishaan-berri@users.noreply.github.com>
Co-authored-by: Ishaan Jaffer <ishaanjaffer0324@gmail.com>
|