* fix(spend-logs): honor store_prompts_in_spend_logs for guardrail_information (LIT-4314)
_get_spend_logs_metadata passed guardrail_information entries through
verbatim, so guardrail hooks that echo the LLM request into
guardrail_response leaked the raw prompt into LiteLLM_SpendLogs.metadata
regardless of store_prompts_in_spend_logs. This mirrored the pre-existing
gap for the other prompt-carrying fields (vector_store_request_metadata,
error_information, etc.), which already sanitize via
_should_store_prompts_and_responses_in_spend_logs.
Add _sanitize_guardrail_information_for_spend_logs alongside the other
per-field sanitizers and wire it into _get_spend_logs_metadata. When the
flag is False the sanitizer replaces guardrail_request and
guardrail_response with REDACTED_BY_LITELM_STRING while preserving every
other typed field on the entry (name, provider, mode, status, timings,
action, violation_categories, risk_score, masked_entity_count, ...) so
guardrail dashboards keep working. When the flag is True (or the field
is None) the entries pass through unchanged.
Widen StandardLoggingGuardrailInformation.guardrail_request from
Optional[dict] to Optional[Union[dict, str]] so the redacted sentinel
satisfies the TypedDict without needing a cast; guardrail_response
already accepted str.
Regression tests cover the three cases (flag=False redacts,
flag=True passes through, None passes through) plus an end-to-end
get_logging_payload path that fails if the wire-in at line 139 is
reverted.
* chore(spend-logs): review nits (one-shot dict build, scrub identifier in tests)
- _redact_prompt_fields_in_guardrail_entry now returns the redacted
dict in one expression instead of seed-then-mutate (TYPE-3)
- swap the illustrative guardrail_name in the new test fixtures for
a generic 'demo-echo-guard' identifier
* chore(spend-logs): only redact guardrail prompt fields when caller supplied them
Greptile P2: the sanitizer was unconditionally writing REDACTED_BY_LITELM
into both guardrail_request and guardrail_response on the copy, so entries
that never carried one of those fields (e.g. a guardrail that only emits
a guardrail_response) came out with a phantom guardrail_request key added.
Guard both assignments with an in-check so the output shape is stable.
Add a mutation-checked regression test that fails if either guard is
removed.
* fix(spend-logs): also redact match_details and classification in guardrail_information
The initial LIT-4314 fix redacted guardrail_request and guardrail_response,
but two other typed fields on StandardLoggingGuardrailInformation also
carry raw prompt content when a first-party guardrail populates them:
- litellm_content_filter/content_filter.py:1676 sets classification =
dict(CompetitorIntentDetection), whose evidence[*].match is a substring
taken directly from the user's normalized prompt (see
litellm_content_filter/competitor_intent/base.py:184-194).
- block_code_execution/block_code_execution.py:571 sets match_details =
guardrail_response = [dict(d) for d in detections], where detections
carry the fenced-code-block content extracted from the user's message.
Reproduced live against localhost:4000 with store_prompts_in_spend_logs
false and a custom guardrail passing tracing_detail with both fields:
before this commit the raw prompt shows up in metadata.guardrail_information[0]
under match_details and classification; after, both are the sentinel.
Widen the two TypedDict fields to Optional[Union[..., str]] so the
sentinel string satisfies the schema without a cast, and consolidate
the redaction set into a tuple so future prompt-carrying additions are
one-line changes.
* fix(spend-logs): normalize non-list guardrail_information shapes in sanitizer
xecguard's logging hook (xecguard.py:246) assigns a bare dict to
standard_logging_object['guardrail_information'] instead of a list,
violating the typed contract Optional[List[StandardLoggingGuardrailInformation]].
Without defensive normalization, _sanitize_guardrail_information_for_spend_logs
iterates the dict's string keys and _redact_prompt_fields_in_guardrail_entry
raises TypeError on {**'guardrail_name'}, which get_logging_payload's
downstream update_database catches with a broad except and silently drops
the entire spend-log write for that request.
Normalize a bare-dict input to a single-item list at the sanitizer's
entry point, and skip any non-dict entries defensively (matching OTEL's
existing isinstance filter at opentelemetry.py:1751-1753 for the same
field). Downstream readers already model this defensively; make the
spend-log write path match.
The root cause is xecguard's writer, not the sanitizer. That is being
tracked as a separate ticket; this PR keeps xecguard-enabled deploys
from silently losing spend logs when store_prompts_in_spend_logs=false.
* fix(types): declare guardrail Union members str-first to avoid poisoning typing cache
CPython's typing module caches Union[...] order-insensitively (first-
construction wins), and litellm/types/utils.py has no 'from __future__
import annotations', so its unions are constructed eagerly at import
time -- before any proxy model. Declaring guardrail_request,
classification, and match_details with dict-first ordering seeds the
typing cache with a dict-first tuple, and later proxy models that
declare custom_llm_provider / model_aliases / vertex_credentials as
Optional[Union[str, dict]] pick up the same dict-first object.
Downstream, Pydantic's get_args() then reports anyOf in dict-first
order, FastAPI emits the OpenAPI accordingly, and 'npm run gen:api'
produces a schema.d.ts diff on unrelated fields, tripping the schema-
sync CI check.
Behaviorally identical in Python and at the wire; the flip only reorders
the union members so the first construction matches how the codebase
had always declared these unions, and 'npm run gen:api' now produces a
zero diff against the committed schema.d.ts.
Follow-up to the full-height sidebar shell. Replaces hand-rolled markup in the sidebar, top bar, and account/usage docks with shadcn primitives so the surface reuses shared components instead of one-off elements
Adds three Base-UI-native primitives that follow the existing conventions (cva + data-slot, no Radix): Meter, Avatar, and Breadcrumb. Badge gains a render prop via Base UI useRender so it can render as an anchor, which activates the variant's existing [a&] styles
Wires the redesigned surface onto them: the version tag is an outline Badge linking to the release notes, the breadcrumb uses the Breadcrumb primitive, the Docs link is a ghost Button, the account initials use Avatar, and the Enterprise usage card uses Collapsible for its open state and Meter for the seat and team gauges. Meter is the correct element for a used-of-total measurement and Base UI ships it natively, so it replaces the hand-rolled bars
Sets the sidebar background to pure white by pointing the --sidebar token at oklch(1 0 0) in light mode, which keeps SidebarUsageCard consistent since it shares the token; the dark block is left untouched
Also repairs layout.test.tsx, which the shell refactor had left red: the dashboard shell now renders DashboardHeader in place of the old navbar, so the test mocks DashboardHeader and asserts on it instead of the navbar the layout no longer mounts
The multi-server list path already relays an upstream 401 from a client-forwarded
server (true_passthrough / oauth_delegate) as an MCPUpstreamAuthError so the caller
re-runs its own upstream OAuth. The single-server REST call path did not: an upstream
401 was masked as a graceful isError result, so an MCP client holding an expired
upstream token never learned it had to re-authenticate
Relay the upstream 401 on the call path too. For these modes the manager calls the
client with raise_on_error=True, extracts the WWW-Authenticate through the existing
upstream-auth exception walk, and raises MCPUpstreamAuthError; the REST endpoint turns
it into a real 401 + WWW-Authenticate. Only 401 is treated as a re-auth signal (a 403 is
a genuine authorization failure that re-auth will not fix, so it stays a masked isError
with a visible warning), matching the list path and MCPUpstreamAuthError's contract. The
legacy oauth2 + delegate_auth_to_upstream mode is deliberately left off the call-path
relay since it is being removed
To keep this expected caller-must-reauth signal from tripping error-rate alerts, the
client layer logs at debug when the caller opted into raise_on_error and therefore owns
the exception (both call_tool/list_tools and the run_with_session helper they share, so an
expected re-auth emits no warning per call either), the manager's non-auth branch logs the
exception type only (never str(e), which for an httpx error embeds the upstream URL a
credential can hide in), and the streamable and REST handlers log the relayed 401 at info
rather than as an error with a traceback
Tests cover the manager raising on a client-forwarded 401 while keeping a 403/503 as a
masked isError, the client-layer debug-vs-error logging split, the streamable handler's
informational isError, and the REST endpoint relaying both the direct and virtual
mcp_tool_call branches as a real 401 + WWW-Authenticate; each was mutation-checked to fail
when the corresponding behavior is broken
Move the admin dashboard to a standard fixed-viewport shell. The sidebar is now full-height with its own scrolling nav (fixed logo header, pinned footer) and the top bar sits only over the content, so the page can no longer scroll past the end of the sidebar
The brand, version, collapse toggle, and account menu move into the sidebar; the AI Gateway/Chat switch, docs/blog/community links, notifications, and worker switcher stay in the top bar. The sidebar is rebuilt on a new shadcn ui/sidebar primitive that uses the existing design-system tokens instead of the antd Menu
This is a pure move-around of the sidebar, header, and content with no behavioral change intended. Chat keeps its own shell and navbar and is deliberately out of scope
Both automated reviewers assumed the semantic filter 400 reaches the
browser as FastAPI's flat detail dict and would render as raw JSON in
the test panel banner. The proxy converts a pre-call hook HTTPException
into a ProxyException that serializes as {"error": {"message": ...}},
which deriveErrorMessage unpacks first; pin that contract with direct
tests
Keep the full provider exception in server-side logs only; the client
receives a fixed actionable message. Also follow implicit exception
context when detecting context window overflows and pin the detection
variants plus the redaction in tests
Resolves LIT-4284
When the embedding model exceeded its context window, the MCP semantic
tool filter silently passed all tools through and reported N->N success
in the filter header; when the overflow happened while embedding tool
descriptions at router build time, the hook was never registered at all
and filtering was silently disabled
Semantic filtering now fails closed on context window overflows: the
request is rejected with HTTP 400 and a message that names the embedding
model and advises switching to one with a larger context window or
disabling the filter. Build time overflows are recorded on the filter so
the hook still registers and blocks MCP tool requests with the same
actionable error while leaving native-only requests untouched. The
dashboard test panel renders the backend message in an error banner
instead of a success state. OpenAI's embedding overflow message
(maximum input length is N tokens) now maps to ContextWindowExceededError
* refactor(ui): colocate the policies view, keeping PolicySelector shared
Group-1 colocation split. The policies folder lived in the shared src/components
dump but is only partly shared: PolicySelector (used by the Playground,
ComplianceUI, and key edit view) and its types stay in @/components/policies,
while the policy-management view (27 files: index, tables, forms, modals,
pipeline builder, and their tests) moves to policies/_components.
Moved files' imports of the retained shared files become @/components/policies
paths; other escaping relative imports are absolutized against the @/ alias
(including @/data/... and the repo test-utils via @/../tests/...), and moved
test files have both their `from` imports and `vi.mock` paths rewritten so the
mocks keep matching the source. Grandfathered lint suppressions for moved files
are re-keyed. PolicySelector did not move, so its external consumers are
untouched. No behavior change.
* fix(ui): keep PolicySelector.tsx eslint suppression at its original path
PolicySelector.tsx stays in src/components/policies (only the management view
moved to _components), but the suppression re-key wrongly moved its
no-nested-ternary entry to the phantom _components path, orphaning the real
file's suppression. Revert that one key. (Greptile P1 on #32720.)
Two of the group-1 colocation splits. Each of these folders lived in the shared
src/components dump but is only partly shared: the page's management view is
segment-owned, while a selector widget is reused by other features. So this
splits them rather than moving wholesale.
tag-management: TagSelector (used by playground) and its types stay in
@/components/tag_management; the management view (index, tag_info, TagTable,
CreateTagModal) moves to tag-management/_components.
vector-stores: VectorStoreSelector (used by organizations and playground) and
its types stay in @/components/vector_store_management; the rest of the
management UI moves to vector-stores/_components.
The moved files' imports of the retained shared files are rewritten to absolute
@/components paths, escaping relative imports are absolutized, and moved test
files have both their `from` imports and `vi.mock` paths rewritten to match.
Grandfathered lint suppressions for moved files are re-keyed. The external
consumers of the selectors are untouched (the selectors did not move). No
behavior change.
Renames each segment's local components/ folder to _components/ (private to the
route, matching Next's _ route-exclusion). Both folders are imported only by
their own page.tsx via the folder index (verified zero external importers
across src, tests, and e2e_tests), so each is a straight rename plus repointing
that one index import; a folder rename keeps every file at the same depth, so
all internal and relative imports are unaffected.
Grandfathered lint suppressions under the two folders (31 entries: cost-tracking
15, prompts 16) are re-keyed to the new paths with counts unchanged. No behavior
change.
Team-level callback_vars (e.g. langsmith_api_key) get spread into
data["metadata"] as four aliases (user_api_key_metadata,
user_api_key_team_metadata, user_api_key_auth_metadata,
user_api_key_auth). When a guardrail hook echoes that metadata into
its guardrail_response, the plaintext credential landed five times
inside LiteLLM_SpendLogs.metadata.standard_logging_guardrail_information[i].guardrail_response
and every downstream sink that reads it (OTel via emit_guardrail_span,
Langfuse, custom loggers).
Add a purpose-built payload walker (mask_credentials_in_payload) that
only masks strings under sensitive-named keys and preserves every
other value (None, ints, floats, bools, tuples, typed objects) verbatim.
The walker reuses SensitiveDataMasker.is_sensitive_key so the pattern
list stays in one place, and unwraps Pydantic models via model_dump()
so nested UserAPIKeyAuth values reached by the walk get scanned as
plain dicts (they are JSON-serialized downstream anyway).
Apply the walker at add_standard_logging_guardrail_information_to_request_data
after the existing secret_fields pop and match/regex redaction, so
every downstream sink sees masked values from a single seam.
At threshold 0.8 azure gpt-realtime fires speech-stop and creates a response, but the committed audio is clipped enough that the response comes back empty (0 transcript, 0 audio), failing the audio-input assertions deterministically. Dropping to 0.5 captures the full utterance so the model produces real content. Verified against a live proxy: 0.8 yields empty responses, 0.5 yields transcript and audio. openai tolerated 0.8; azure did not
azure batch used azure/gpt-4.1-mini-batch; gpt-4.1-mini is deprecating (2026-11-04)
and can no longer be deployed, so point it at gpt-5.4-mini (Global Batch) and bump
the api_version to 2025-04-01-preview. Requires an Azure Global Batch deployment
named gpt-5.4-mini-batch plus AZURE_API_BASE/AZURE_API_KEY on the proxy.
xai/grok-4-1-fast-non-reasoning is deprecated (2026-05-15); update the commented
xai realtime provider and the coverage-matrix doc to xai/grok-4-1-fast.
Addresses three issues in the migrated Team Info virtual keys table, all
pre-existing behavior carried over from the tremor version:
- Changing the sort now resets to page 1. Previously handleSortingChange
routed through handleFilterChange with skipDebounce=true, which skipped the
pageIndex reset, so sorting while on a later page asked the server for that
page of the newly sorted results (an arbitrary slice).
- Reset Filters now restores the default sort. It previously reset the filter
fields and page but never touched the sorting state that actually drives the
query, so the sort indicator and server order persisted.
- Removes the dead Sort By / Sort Order keys from the filters object; sort is
derived solely from the sorting state, so those keys were written but never
read. Sort now lives in one place.
Adds regression tests for the page-reset-on-sort and sort-reset-on-filter-reset
behaviors (both fail if either fix is reverted).
Colocation follow-up to the App Router migration: move each page's owned
components out of the shared src/components dump and into its route segment's
_components/ folder, draining the shared bucket. Convention: a component used
by exactly one segment goes in that segment's _components/ (private, matching
Next's _ route-exclusion); a component shared by 2+ segments stays in
@/components. No new _shared/ folder.
Rename-in-place (segment already had a local components/ folder):
- api-reference (also relocates the shared CodeBlock, used by playground and
cost-tracking, to @/components/CodeBlock)
- memory, budgets, access-groups
- caching, projects, guardrails-monitor
Extract from src/components (page view lived in the shared dump):
- AdminPanel -> admin-panel, organizations -> organizations,
general_settings -> router-settings, usage -> old-usage
Each folder/view was verified to have no importer other than its own page
(cross-checked across src, tests, and e2e_tests). Relative imports inside moved
single files are rewritten to absolute @/components/*; colocated tests move with
their subject and have their vi.mock paths rewritten to match. Grandfathered
lint suppressions (tremor, react-hooks, and similar, all pre-existing) are
re-keyed to the new paths with counts unchanged. No behavior change.
Drop the duplicate local DEFAULT_PAGE_SIZE_OPTIONS in DataTable.tsx and import
the one already exported from DataTablePagination.tsx, removing the divergence
risk if the canonical list changes.
Follow-up polish on the migrated Team Info virtual keys table: widen the Key
ID column by 20px (100 -> 120), nearly double Created By (70 -> 130) so the
name and popover fit, and remove the Last Active header info icon (and its now
unused InfoCircleOutlined import).
The preview endpoint infers client_credentials when the inherited client_id, client_secret, and
token_url are all present (common once DCR or discovery filled them) and then strips the forwarded
bearer to preview as M2M, so the staged interactive token was silently unused; sending
oauth2_flow=authorization_code bypasses the inference. spec_path now rides along so OpenAPI servers
take the spec-based preview path the create form gets. clearHeldOAuthToken also empties the tool
list, mirroring the create form's clearTools, so a preview fetched with the discarded token never
lingers while the refetch is in flight
The per-row delete_many loop becomes a single delete filtered to the enumerated OAuth users'
(user_id IN, server_id) pairs; same rows deleted, same BYOK-sparing precision, same count-mismatch
detection, one round-trip instead of N
The field doc still said scheme + host + path while the redactor now strips the path along with
userinfo, query, and fragment, since hosted MCP servers routinely embed the credential in the path
The create and edit submit paths for true_passthrough and oauth_delegate persist only the tool
configuration: the parametrized create test authorizes, disables the allowlist, and asserts nothing
is persisted before submit, then that the create payload carries allowed_tools but no credentials
and no occurrence of the token anywhere in the serialized payload, no per-user DB credential is
written, and the token is committed to sessionStorage only, keyed to the created server. The edit
save test gains the same serialized-payload assertion
For authorization_code the edit preview listed tools by server_id only, relying on the stored
per-user DB credential, so a token authorized in the edit session gave an empty preview until the
admin saved; the create form previews the identical state through the config-based preview
endpoint, which takes the token explicitly. The edit fetch now routes through that same endpoint
when a staged interactive token is held, built from the form values with the saved record as
fallback, and keeps the by-server_id listing for every other case
The identity used to pick the audience from spec_path only when
values.transport was OPENAPI, but the create form keeps transport in component
state rather than form values, so spec_path edits on OpenAPI servers never
invalidated a held token. Comparing url and spec_path independently mirrors
the backend's mcp_oauth_token_identity and fires regardless of whether
transport is present. Invalidation now also wipes only credentials; the
admin-typed endpoint fields are kept
LiteLLM_MCPUserCredentials stores BYOK API keys in the same column as per-user
OAuth tokens, so the purge on a mint-relevant config change now deletes only
rows whose payload decodes as an OAuth2 credential, each by its
(user_id, server_id) pair, instead of every row for the server. An api_key
server whose url changes purges nothing. delete_mcp_server now also
invalidates each enumerated user's cached token so a re-created server reusing
the id cannot serve tokens minted for the deleted one, and both cache drops
are best-effort
The staged access token never reaches formValues (it is not a registered form field), so the
assertion could not fail; the DCR client pair is the leak the test actually pins, proven by the
mutation run
The snapshot read only feeds the stale-token purge decision; leaving it unguarded meant a failed
read would 500 an edit whose update would have succeeded, and it broke
test_edit_mcp_server_redacts_credentials, whose mocked prisma is not awaitable on the un-patched
get_mcp_server path. A failure now logs and skips the purge, consistent with the purge half already
being best-effort. Adds the first endpoint-level coverage of the edit purge wiring: purge on a
mint-relevant change, no purge when the identity is unchanged, and edit success with purge skipped
when the snapshot read raises
The purge takes an injectable invalidate_token_cache callable defaulting to the manager's shared
invalidation, and MCPServerManager takes an injectable per_user_token_cache alongside the existing
per_user_oauth_token_store, so tests inject fakes instead of monkeypatching the global manager and
the module-level cache. The new identity helpers drop Any for object throughout
Review follow-ups on the stale-token invalidation. The backend identity now decrypts client_id and
client_secret before comparing: the stored values are NaCl-encrypted with a fresh nonce on every
write, so comparing ciphertext flagged every routine save as a mint-relevant change and purged
per-user tokens that were still valid. The identity also gains spec_path, the audience for OpenAPI
servers, and parses credentials stored as a JSON string
The purge now routes each (user, server) through the manager's invalidate_user_oauth_token_cache,
which becomes the single invalidation point covering both the legacy per-user token cache and the
v2 per-user OAuth token store; previously the purge evicted only the legacy cache while the revoke
path evicted only the v2 store, so each path left the other cache serving a replaced token until
its TTL. A credential row racing in between the find and the delete is now detected via the
delete_many count and logged; its cache entry expires by TTL
On the dashboard, CLEARED_ON_INVALIDATION and the staleness check move to types.tsx as the single
shared implementation for both forms. The edit form's transport handler now rechecks the identity
after its programmatic setFieldsValue calls, which antd does not report through onValuesChange, so
a token no longer survives a transport switch that clears the mint target. The create form rebuilds
formValues from the post-reset form state after an invalidation instead of publishing the pre-reset
snapshot, so the tool preview can no longer refetch with the discarded DCR client. Both transport
handlers now share the recheck, which also stops the create form from over-invalidating on an
http to sse swap that keeps the same url and therefore the same audience
An admin who ran Authorize & Fetch and then changed a field that determines which upstream OAuth
token gets minted kept using the stale token for tool preview, sessionStorage, and (on the backend)
the stored per-user credential and its cache. Grounded in RFC 8707/8693 and the MCP auth spec, a token
is bound to one tuple: resource/audience (url), OAuth mode/grant (auth_type, oauth_flow_type), the
authorization-server endpoints, and the OAuth client + scopes. A shared getOAuthAuthorizationIdentity
captures exactly those fields; transport (http/sse on the same url is the same audience) and
delegate_auth_to_upstream (a downstream-usage toggle never sent to the authorize request) are excluded.
UI: both the create and edit forms now discard the held token (React state / sessionStorage / hook,
plus the fetched token + DCR client in form.credentials) whenever the identity diverges from the one it
was authorized against, re-applying the admin's in-flight edit so it is never wiped. The check lives in
one shared helper so the two forms cannot drift.
Backend: editing an MCP server now compares the pre/post identity and, on a mint-relevant change, purges
every stored per-user OAuth credential for the server (DB row + per-user token cache) so no user
forwards a token minted for a resource/AS/client that no longer matches. Best-effort; a purge failure
never fails the update.
Second proof-of-concept consumer for the shared DataTable. Replaces the
hand-rolled tremor table in the Team Info Virtual Keys tab with DataTable in
server-sort and server-pagination mode plus column resizing; the file drops
about 150 lines. Sortable headers now use DataTableSortHeader, pagination is
a detached DataTablePagination driven by the page state, the id-cell still
opens the key drawer, and the body scrolls under a sticky header via
maxBodyHeight. Two behavior changes: the pagination control is the
standardized bar (row range plus page-size select) rather than the old
Previous/Next buttons, and a sort header cycles ascending/descending without
a third unsorted state, which also removes a latent case where clearing the
sort left the server sorted.
Updates the TeamVirtualKeysTable and TeamInfo tests to the new pagination,
adds a test that a sort-header click routes to useKeys as a server sort, and
lowers the no-large-inline-object-arg metric by one and the file's
no-nested-ternary suppression from two to one to match the leaner code.
