The tests were mocking `filter_server_ids_by_ip` but the production
code in server.py now calls `filter_server_ids_by_ip_with_info` which
returns a (server_ids, blocked_count) tuple. Update all 8 mock sites
to use the correct method name and return signature.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When calling non-text-embedding-3 models routed through the openai provider
(e.g. nvidia/llama-3.2-nv-embedqa-1b-v2), passing `dimensions` previously
raised an UnsupportedParamsError unconditionally. This fix threads
`allowed_openai_params` through the embedding call stack so that providers
can opt-in to passing `dimensions` by including it in the list.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The old test assumed ArizePhoenixLogger reused the global TracerProvider.
With the nested traces fix, Phoenix now creates its own dedicated provider
and produces litellm_proxy_request + litellm_request + raw_gen_ai_request
spans independently.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- ArizePhoenixLogger now creates spans on its own dedicated TracerProvider
instead of trying to reuse parent spans from the global otel TracerProvider
(which were invisible in Phoenix since they go to a different exporter)
- Auto-initialize ArizePhoenixLogger when otel callback is configured and
Phoenix env vars (PHOENIX_API_KEY, PHOENIX_COLLECTOR_*) are detected
- Use exact type check in get_custom_logger_compatible_class to prevent
ArizePhoenixLogger (subclass) from being returned when looking up otel
- Fix tool_permission guardrail to check non-function tools like
code_interpreter (previously skipped with `type != "function"`)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* feat(realtime guardrails): end_session_after_n_fails + Endpoint Settings wizard step
Adds per-session violation thresholds and an optional endpoint-settings step
to the guardrail wizard for /v1/realtime.
Backend:
- Add end_session_after_n_fails, on_violation, realtime_violation_message fields
to BaseLitellmParams (no DB migration — stored in existing JSON column)
- Store same fields on CustomGuardrail instance attrs
- Pass through in litellm_content_filter initializer
- Track _violation_count per RealTimeStreaming session; close backend_ws when
on_violation=end_session OR violation count >= end_session_after_n_fails
- Use realtime_violation_message as the spoken text (falls back to guardrail
error string if not configured)
UI (add_guardrail_form.tsx):
- Rename "Default Categories" step to "Topics"
- Add step 5 "Endpoint Settings (Optional)" for content filter guardrails
- Call type dropdown shows /v1/realtime
- Settings are in a collapsed accordion (closed by default)
- "End session after X violations" + on_violation radio + spoken message field
Tests: 2 new tests in test_realtime_streaming.py
- test_end_session_after_n_fails_closes_connection
- test_on_violation_end_session_closes_on_first_fail
* fix(test): move inline imports to module level in realtime streaming tests
* Update ui/litellm-dashboard/src/components/guardrails/add_guardrail_form.tsx
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
---------
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
* fix(realtime): guardrails with pre_call/post_call mode now work on realtime WebSocket; return error directly to consumer
* fix(realtime guardrails): address code review feedback
- Restore session.update injection for audio/VAD path, but only when
realtime_input_transcription guardrails are configured (not pre_call).
Forward session.created to the client first so no error arrives before
the client sees the session.
- Change _swallow_next_response_create bool to int counter so consecutive
blocked items are handled correctly.
- Extract _build_litellm_metadata() helper to eliminate duplicated
metadata-building logic across OpenAI/Azure/XAI provider branches.
- Plumb litellm_metadata and user_api_key_dict to Azure and XAI handlers
so guardrails work for those providers too.
- Add tests for session.update injection, no-inject for pre_call-only,
and consecutive-block counter.
* simplify: remove response.create swallowing after guardrail block
When an item is blocked, the error event is already sent to the client.
The subsequent response.create from the client is fine to forward through —
the LLM may respond to previous context which is acceptable behavior.
Removing the swallow counter eliminates unnecessary state tracking.
* fix(adapter): populate cache_read_input_tokens from prompt_tokens_details
The Anthropic adapter's translate_openai_response_to_anthropic checked
only the private _cache_read_input_tokens attr (set by Anthropic/DeepSeek)
but not prompt_tokens_details.cached_tokens (set by OpenAI/Azure).
Use prompt_tokens_details.cached_tokens directly — it is already extracted
and is the standard field populated by all providers.
Fixes#22089
* fix(adapter): apply same cache_read_input_tokens fix to streaming path
The streaming path in translate_streaming_openai_response_to_anthropic
had the same bug — relying on _cache_read_input_tokens instead of
prompt_tokens_details.cached_tokens.
* fix(proxy): improve auth exception logging levels and add structured context
Downgrade expected auth failures (ProxyException, HTTPException < 500,
BudgetExceededError) from ERROR to WARNING log level to reduce noise from
routine rejected requests (e.g. missing/invalid API keys on polled endpoints
like /schedule/model_cost_map_reload/status).
Unexpected exceptions and HTTPException with status >= 500 still log at
ERROR with full traceback.
Enrich log messages with structured context: route, HTTP method, masked
API key (using existing abbreviate_api_key), error type, and error code.
All fields also passed via log extra dict for log aggregation tools.
Fixes#21293
* Update tests/test_litellm/proxy/auth/test_auth_exception_handler.py
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
---------
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
* feat(vertex_ai): add Vertex AI Gemini Live support via unified /realtime endpoint
Adds VertexAIRealtimeConfig which translates the OpenAI Realtime WebSocket
protocol to Vertex AI BidiGenerateContent. Supports voice in/voice out
(16 kHz mic → 24 kHz speaker) and text in/text out through the proxy's
/realtime endpoint.
Key changes:
- New litellm/llms/vertex_ai/realtime/transformation.py with VertexAIRealtimeConfig
- Builds correct wss:// URL (regional + global)
- OAuth2 Bearer token auth (not API key)
- Full model path (projects/.../publishers/google/models/...)
- Ignores session.update (Vertex AI only accepts one setup message)
- realtime_api/main.py: vertex_ai branch resolves OAuth token + constructs config
- llm_http_handler.py: auto-sends session setup before bidirectional_forward
- gemini/realtime/transformation.py: fix crashes on empty turnComplete events
- realtime_streaming.py: try/except guard so bad messages don't kill the loop
- proxy_server.py: add missing websockets.exceptions import
* docs: add vertex_realtime to sidebars
* fix: drop unknown event types in Gemini transform; add vertex_ai health check
* fix: propagate UUID fallback IDs from transform_content_done_event to return_additional_content_done_events
* fix: route guardrail backend sends through provider transform; fix str.strip misuse for model prefix
* fix: handle Vertex AI full resource path in session.created; route guardrail block sends through _send_to_backend
* fix: remove unused VertexBase in transformation.py; apply UUID fallback in return_additional_content_done_events
Adds a new block_code_execution guardrail that detects markdown fenced code blocks
in request/response content and blocks or masks them by language. Includes full
UI integration, type definitions, compliance test dataset, and 26 unit tests.
Key guardrail capabilities:
- Regex-based fenced code block detection with configurable blocked languages
- Confidence scoring with tunable threshold
- Execution-intent heuristics (request-side only) with conflict resolution
- Block or mask actions for detected code
- Support for pre_call, post_call, and during_call event hooks
Security hardening:
- Response-side blocking skips intent heuristics (LLM output doesn't contain
user intent phrases, so checking would silently disable post_call blocking)
- No-execution short-circuit includes conflict resolution: if both no-execution
and execution phrases match, execution intent wins
- Tightened overly broad phrases to prevent trivial bypass
- _normalize_escaped_newlines only applies to pure-escaped payloads to avoid
corrupting content that discusses escape sequences
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Add post-call hook for Lakera guardrail and mask PII in responses
* Add post-call hook for Lakera and mask PII in responses
* Fix post-call hook: pass event_type to call_v2_guard
* Address Greptile review: return ModelResponse, fix mutation, add header, test location, mask order
- PII masking path: return ModelResponse instead of dict so deployment hook accepts it
- Avoid mutating request data: deep copy original_messages and messages in _mask_pii_in_messages
- Add guardrail header in PII-only return path
- Add test in tests/test_litellm/ (test_lakera_ai_v2.py) per PR checklist
- Sort PII payload spans by (start,end) descending so multiple spans in one message mask correctly
Co-authored-by: Cursor <cursoragent@cursor.com>
* Updated ponteital for index mismatch when choices have null content and inconsistent on_flagged access pattern
* Update litellm/proxy/guardrails/guardrail_hooks/lakera_ai_v2.py
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
* Update to explicitly state supported endpoints - chat completions
* Fix minor lint error on masked_entity_count
---------
Co-authored-by: Steve <steve.giguere@lakera.ai>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Backend - Spend Log Storage for Realtime Calls:
- Collect user voice transcripts and text input during WebSocket sessions
- Store collected messages in spend logs when store_prompts_in_spend_logs enabled
- Capture tool definitions from session.update and tool calls from response.done
- Enrich proxy_server_request with tools and response with tool_calls for UI
Backend - WebSocket Auth:
- Support browser-based auth via Sec-WebSocket-Protocol subprotocol
- Echo back subprotocol on WebSocket accept
UI - Realtime Playground:
- New RealtimePlayground component with WebSocket voice+text chat
- Mic recording (PCM16 24kHz), server VAD, audio playback, text input
- Handle binary WebSocket frames (Blob/ArrayBuffer decoding)
- Add /v1/realtime endpoint option to playground endpoint selector
UI - Tools Section for Realtime Logs:
- Extract tool calls from realtime response format (response.tool_calls
and response.results[].response.output[].type=function_call)
Tests:
- 15 new backend tests for realtime streaming and spend log storage
- 4 new UI tests for realtime tool call extraction
Fixes pre-existing build errors:
- ToolPolicies.tsx: duplicate import, antd styles type
- create_key_button.tsx: missing message import
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
- Add per-command error check in _pipeline_lpop_helper to match _pipeline_rpush_helper, preventing silent data loss on WRONGTYPE errors
- Fix pre-existing bug: org spend queue metric was using REDIS_DAILY_SPEND_UPDATE_QUEUE instead of REDIS_DAILY_ORG_SPEND_UPDATE_QUEUE
- Add test for per-command LPOP pipeline error propagation
* feat(proxy): add max_iterations limiter for agent session loops (#22058)
Adds a new proxy hook that enforces a per-session cap on the number of
LLM calls an agentic loop can make. Callers send a session_id with each
request, and the hook counts calls per session, returning 429 when the
configured max_iterations limit is exceeded.
- Uses Redis Lua script for atomic increment (multi-instance safe)
- Falls back to in-memory cache when Redis unavailable
- Follows parallel_request_limiter_v3 pattern
- Configurable via key metadata: {"max_iterations": 25}
- Session counters auto-expire via TTL (default 1hr)
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add new code execution dataset
* feat(agent_endpoints/): allow giving agents keys
* fix: ui fixes
* feat: allow assigning mcp servers to agents
* fix: eliminate duplicate DB queries in MCP agent auth and N+1 in agent listing (#22110)
- Extract _get_agent_object_permission helper so _get_allowed_mcp_servers_for_agent
and _get_agent_tool_permissions_for_server share a single DB fetch instead of
each independently querying the same agent row (was 1+N queries per MCP request)
- Use include={"object_permission": True} on find_many in get_all_agents_from_db
to eagerly load permissions in one query instead of N+1
- Use include={"object_permission": True} on create/update/find_unique in all
agent CRUD operations, removing attach_object_permission_to_dict follow-up calls
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Failure spend logs were missing key metadata (key alias, user ID, team ID,
team alias) in two scenarios:
1. Auth errors (401 ProxyException): auth_exception_handler creates a
minimal UserAPIKeyAuth with only api_key and request_route set — all
other fields are null. The failure hook now looks up the full key object
from cache/DB using the key hash to populate the missing fields.
2. Post-auth failures (provider errors, rate limits): key fields are
present but team_alias is always null because LiteLLM_VerificationTokenView
SQL view does not include team_alias. The failure hook now looks up the
team object from cache to populate team_alias.
Both lookups are non-fatal and wrapped in try/except.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Google's Interactions API spec changed the status enum:
- Values are now lowercase (was uppercase)
- 'UNSPECIFIED' value was removed
Updated test to match the current spec from:
https://ai.google.dev/static/api/interactions.openapi.json
Add a `request_duration_ms` column to `LiteLLM_SpendLogs` to track request
duration. New rows are computed at write time. Legacy rows use a COALESCE
fallback in the `/spend/logs/ui` query to compute duration on the fly from
`endTime - startTime`. The field is also sortable in the UI endpoint.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* feat(agents): assign virtual keys to agents
- Add agent_id field to LiteLLM_VerificationToken (schema.prisma + _types.py)
- Pass agent_id through key generation endpoint so keys can be scoped to an agent
- Refactor Add Agent wizard to 3-step flow (Configure → Assign Key → Ready)
- Configure: all agent fields, custom/other type with just name+description
- Assign Key: create new key or reassign existing key to agent
- URL is now optional for easy discovery
- Add "Agent" ownership option to Create Key modal on Virtual Keys page
with agent selector dropdown
- Extract CreatedKeyDisplay into shared component, reused in both flows
- Add keyCreateForAgentCall networking helper
- Add test for agent_id key generation
* fix(agents): code quality fixes from self-review
- Fix test_generate_key_helper_fn_agent_id: remove bare except clause,
use explicit assert mock_insert.called, use .kwargs for clean arg access
- Remove no-op conditional in handleNext (both branches were identical)
- Validate selectedExistingKey before calling keyUpdateCall
- Validate selectedAgentId before setting on formValues in create_key_button
* fix(ui): replace deprecated Tremor Button with Ant Design Button in CreatedKeyDisplay
These params were silently dropped for Chat Completions because they
were missing from the supported params whitelist. Also adds
prompt_cache_retention to the Responses API TypedDict and fixes
misleading cache_control comments in OpenAI prompt caching docs.
* Add Ask AI chat component to Usage page
- Create UsageAIChatModal component with streaming chat interface
- Integrate with existing model hub for model selection
- Pass usage data context (spend, models, providers, keys) to AI
- Add Ask AI button next to Export Data button in global view
- Add tests for the new component and integration
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Convert Ask AI from modal to right-side sliding panel
- Replace UsageAIChatModal with UsageAIChatPanel
- Panel slides in from right side, usage page stays visible
- Full-height panel with header, model selector, chat area, and input
- Smooth CSS transition for open/close animation
- Update tests for new panel component (34 tests passing)
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Remove build output directory from tracking
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Add backend AI usage chat endpoint with tool calling
Backend:
- New /usage/ai/chat SSE streaming endpoint
- AI agent has get_usage_data tool that queries /user/daily/activity/aggregated
- Follows same architecture as policy AI suggest (litellm.acompletion + tools)
- Non-admin users are restricted to their own data
- 12 backend unit tests
Frontend:
- Panel now calls /usage/ai/chat backend endpoint via SSE
- Removed direct OpenAI client calls from frontend
- Added usageAiChatStream networking function following enrichPolicyTemplateStream pattern
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Make model selection optional, default to gpt-4o-mini on backend
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Add team/tag tools, status indicators, and improved AI agent
- AI agent now has 3 tools: get_usage_data, get_team_usage_data, get_tag_usage_data
- Stream status events (Thinking... Fetching... Analyzing...) to UI
- Frontend shows spinner + status text during tool execution
- Better system prompt guiding tool selection
- Entity summariser for team/tag data with ranked breakdowns
- 13 backend tests, 34 frontend tests passing
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Fix: inject today's date into system prompt so AI resolves relative dates correctly
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Show tool calls as distinct steps + render markdown in responses
- Backend emits tool_call events with tool_name, label, args, and status
- Frontend shows each tool call as a step with ✓/spinner/✗ indicator
- Tool call steps show icon, label, date range, and filters
- AI responses rendered with ReactMarkdown (bold, lists, tables, code)
- Cursor-like UX: Thinking → tool calls → Analyzing → streamed answer
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Refactor backend for code quality: proper types, constants, all functions ≤50 LOC
- TypedDict for SSE events (SSEStatusEvent, SSEToolCallEvent, etc.) and ToolHandler
- Constants for table names, entity fields, temperature, page sizes, top-N limits
- Shared _query_activity() eliminates duplicated fetch logic
- _accumulate_breakdown() + _ranked_lines() replace inline aggregation loops
- Extracted _process_tool_call() and _stream_final_response() from main stream fn
- Black + Ruff clean, all 15 functions verified ≤50 LOC
- Replaced Tremor Button with Antd Button in panel (Tremor deprecated per AGENTS.md)
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Address greptile review: security fixes and input validation
- Restrict team/tag tools to admin-only users (non-admins only get get_usage_data)
- Constrain ChatMessage.role to Literal['user', 'assistant'] to prevent system prompt injection
- Add test for base tools restriction (non-admin gets 1 tool, admin gets 3)
- Issues 3 (unused imports) and 4 (inline datetime) were already fixed in prior commit
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Address greptile round 2: sanitize errors, defense-in-depth allowlist, revert tsconfig
- Sanitize error messages: generic 'An internal error occurred' sent to client,
full exception logged server-side via verbose_proxy_logger
- Defense-in-depth: _process_tool_call validates fn_name against role-based
allowlist before dispatch (even though LLM only receives allowed tools)
- Revert tsconfig.json jsx back to 'preserve' (Next.js recommended default)
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Role-scoped system prompt + additional test coverage
- System prompt is now role-aware: admin sees all 3 tool descriptions,
non-admin only sees get_usage_data (consistent with tool filtering)
- Added tests: non-admin prompt excludes team/tag tools, date injection
- 15 backend tests, 34 frontend tests passing
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* Fix LLM arg validation + cap conversation size at 20 messages
- _resolve_fetch_kwargs uses .get() with ValueError for missing dates
(handles malformed LLM tool arguments gracefully)
- MAX_CHAT_MESSAGES = 20 constant; backend truncates to last 20
- Frontend also sends only last 20 messages per request
- Prevents excessive token usage and context-length errors
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
_cached_get_model_group_info uses @lru_cache but had no invalidation,
causing stale model group info (TPM/RPM limits) after dynamic deployment
changes. Add cache_clear() at all 5 model_list mutation sites.
* feat(proxy): tool policies - auto-discover tools, manage policies, guardrail enforcement
- New LiteLLM_ToolTable in schema.prisma to store discovered tools
- Auto-discovery: tools seen in LLM responses get upserted via ToolDiscoveryQueue
(hooks into DBSpendUpdateWriter, same pipeline as spend tracking)
- Management endpoints: GET /v1/tool/list, GET /v1/tool/{name}, POST /v1/tool/policy
- ToolPolicyGuardrail: blocks tool_calls in responses based on policy setting
- UI: Tool Policies page under Guardrails section with policy selector,
filters by policy/team/key, live tail, sortable table
- Unit tests for queue, writer, endpoints, guardrail
* feat(tool-policies): track call_count + discover tools from request body and /messages API
- Add call_count column to LiteLLM_ToolTable; incremented on every flush
- Extract tools from request body too (not just response tool_calls):
- OpenAI /chat/completions: tools[].function.name
- Anthropic /messages pass-through: request_body.tools[].name
- Show call_count column in UI table (sortable)
- UI: drop dual_llm option, keep only trusted/blocked
* fix: address greptile review feedback
- Remove redundant @@index([tool_name]) from schema.prisma (tool_name has @unique which already creates an index)
- Replace gen_random_uuid()::text with str(uuid.uuid4()) for portability
- Rewrite test_tool_registry_writer.py to mock execute_raw/query_raw (actual implementation) instead of Prisma model methods
- Fix test patches in test_tool_management_endpoints.py to target source modules since imports are inside function bodies
- Add "Tool Policies" page title to ToolPolicies.tsx
* fix: address greptile review round 2
- Replace NOW() with Python datetime parameter in tool_registry_writer (SQLite portability)
- Fix cache key collision in tool_policy_guardrail: use null-byte separator instead of colon
- Remove type==function filter from request-side tool extraction to match response-side behavior
- Clear seen_tool_names on flush so call_count increments per batch cycle not per pod lifetime
* fix: address greptile review round 3
- Fix test_seen_names_persist_across_flushes to match actual per-flush-cycle behavior
- Update module docstring in tool_discovery_queue.py to accurately describe flush behavior
- Add created_at/updated_at to raw SQL INSERT in batch_upsert_tools and update_tool_policy
* fix: cache tool policies per tool name not per combination
Previously the cache key was built from the full set of tool names in a
request, so each unique combination of tools got its own cold cache entry
and triggered a separate DB query. With N distinct tools across requests
this was effectively a DB hit on every request.
Now each tool name is cached individually. Cache hits are checked per
tool, only missing tools are fetched from DB in a single batch query,
and each result is cached separately. Once a tool's policy is warm,
any subsequent request using that tool benefits from the cache regardless
of what other tools are in the request.
* Update ui/litellm-dashboard/src/components/ToolPolicies.tsx
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
---------
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Avoid constructing a default Usage() object that gets immediately
overwritten by convert_to_model_response_object. Set usage=None
instead; the real Usage is assigned via setattr later.
Also fix Bedrock Qwen2/Qwen3 transform_response to assign a new
Usage object instead of mutating a potentially missing one.
* fix(router): preserve _hidden_params in FallbackStreamWrapper so x-litellm-overhead-duration-ms is emitted for streaming requests
* test(router): add regression test for FallbackStreamWrapper _hidden_params preservation
