Open Source Dashboard - Giulio Leone

langchain-ai/langchain128k

fix(core): prevent RunnableRetry.batch output corruption when partial failures occur

Python4 giorni fa

1752

langchain-ai/langchain128k

fix(openai): prevent crash when disable_streaming="tool_calling" in streaming mode

Python4 giorni fa

2223

langchain-ai/langchain128k

fix(core): pass bound tools to token counter in trim_messages

Python3 giorni fa

1573

langchain-ai/langchain128k

fix(core): make trim_messages respect tool-call group integrity

Python4 giorni fa

51424

open-webui/open-webui126k

fix(config): RAG_OPENAI env vars now correctly inherit OPENAI_API values

Python4 giorni fa

62

open-webui/open-webui126k

fix(config): RAG_OPENAI env vars now correctly inherit OPENAI_API values

Python4 giorni fa

62

browser-use/browser-use79k

fix: use tempfile.gettempdir() instead of hardcoded /tmp for Windows compat

Python5 giorni fa

55

run-llama/llama_index47k

fix(openai): exclude parallel_tool_calls from request when no tools provided

Python5 giorni fa

64

crewAIInc/crewAI45k

fix(flow): allow Flow subclasses in crew.copy() for kickoff_for_each (fixes #4555)

Python4 giorni fa

1634

crewAIInc/crewAI45k

fix(crew): preserve full task output in crew.kickoff() (fixes #4603)

Python4 giorni fa

3361

crewAIInc/crewAI45k

fix(cli): memory leak in Repository.is_git_repo() due to @lru_cache

Python4 giorni fa

406

agno-agi/agno38k

fix: wrap input as list for non-default Azure models

Python5 giorni fa

1866

agno-agi/agno38k

fix: prevent parameter name collision between framework and MCP tool args

Python4 giorni fa

36617

agno-agi/agno38k

fix: parse raw JSON before cleaning to preserve fenced code blocks (fixes #5901)

Python4 giorni fa

12313

agno-agi/agno38k

fix: add max_history_tokens to prevent session overflow (fixes #5011)

Python4 giorni fa

2341

agno-agi/agno38k

fix: resolve output_schema and reasoning conflict for Anthropic (fixes #5582)

Python4 giorni fa

215

agno-agi/agno38k

fix: prevent deadlock in parallel MCP tool calls with header_provider (fixes #6094)

Python4 giorni fa

392125

agno-agi/agno38k

fix(litellm): prevent empty tool name from overwriting valid name in streaming

Python5 giorni fa

373

BerriAI/litellm38k

fix(tools): gracefully repair truncated JSON in tool call arguments

Python4 giorni fa

23712

BerriAI/litellm38k

fix: WebSearch interception fails with thinking enabled + SpendLog dedup

Python2 giorni fa

51722

BerriAI/litellm38k

fix(streaming): prevent Vertex AI Claude content truncation when finish_reason races content

Python2 giorni fa

991

BerriAI/litellm38k

fix(featherless_ai): use correct FEATHERLESS_AI_API_KEY env var name

Python4 giorni fa

453

BerriAI/litellm38k

fix(ui): parse YAML blocklist file and populate blocked words

Python5 giorni fa

9,24315

BerriAI/litellm38k

fix(proxy): bound spend_log_transactions to prevent OOM (#21514)

Python4 giorni fa

25550

BerriAI/litellm38k

fix(cost): fix Vercel AI Gateway cost tracking returning $0

Python4 giorni fa

14415

BerriAI/litellm38k

fix(adapter): clamp max_tokens on Anthropic pass-through to backend model limit

Python4 giorni fa

2151

BerriAI/litellm38k

fix(router): resolve Gemini CLI model variant suffixes (-customtools)

Python4 giorni fa

4502

BerriAI/litellm38k

fix(cost): audio transcription cost always 0 for custom-priced models

Python4 giorni fa

349

BerriAI/litellm38k

fix(proxy): add default max_idle_connection_lifetime to prevent stale PostgreSQL connections

Python4 giorni fa

121

BerriAI/litellm38k

fix(proxy): persist default budget_id for implicitly created end users

Python4 giorni fa

3251

BerriAI/litellm38k

fix(anthropic-passthrough): defer first content_block_start to respect thinking block ordering

Python4 giorni fa

49151

BerriAI/litellm38k

fix(router): downgrade Azure model identification from ERROR to DEBUG

Python4 giorni fa

1903

BerriAI/litellm38k

fix(http): add litellm.aclose() and sync connector cleanup for session leaks

Python4 giorni fa

1141

BerriAI/litellm38k

fix(logging): prevent RuntimeError when event loop changes across tests

Python4 giorni fa

40354

BerriAI/litellm38k

fix(alerting): prevent false 'Requests are hanging' Slack alerts

Python4 giorni fa

3485

BerriAI/litellm38k

fix(proxy): strip null bytes from spend log data before PostgreSQL insertion

Python4 giorni fa

127

BerriAI/litellm38k

fix(streaming): propagate provider cost through streaming chunk builder

Python4 giorni fa

2192

BerriAI/litellm38k

fix(proxy): remove wildcard patterns from /models endpoint response

Python4 giorni fa

4107

BerriAI/litellm38k

fix(bedrock): normalize double-wrapped embedding input to prevent JSONArray error

Python4 giorni fa

2336

BerriAI/litellm38k

fix(streaming): thread-safe model_copy to prevent dictionary changed size error

Python4 giorni fa

2669

BerriAI/litellm38k

fix(db): fix DailyTagSpend view double-counting with multi-tag requests

Python4 giorni fa

15641

BerriAI/litellm38k

fix(cost): fix rerank cost calculation always returning 0

Python4 giorni fa

9916

BerriAI/litellm38k

fix(logging): enqueue async callbacks synchronously to prevent task cancellation (#8842)

Python4 giorni fa

13041

BerriAI/litellm38k

fix: merge x-litellm-tags with team/project tags instead of overwriting (fixes #14052)

Python4 giorni fa

1324

BerriAI/litellm38k

fix: preserve 400 status code for context window errors in streaming (fixes #18689)

Python4 giorni fa

1172

BerriAI/litellm38k

fix(logging): stop configuring root/external loggers (fixes #16284)

Python4 giorni fa

21231

BerriAI/litellm38k

fix(proxy): prevent OOM on Virtual Keys UI with large key counts (#19477)

Python4 giorni fa

47189

BerriAI/litellm38k

fix(http): add connection pool limits to prevent file descriptor exhaustion (fixes #13220)

Python4 giorni fa

2166

BerriAI/litellm38k

fix(otel): decouple OTel tracing from proxy dependencies (fixes #13081)

Python4 giorni fa

29564

BerriAI/litellm38k

fix(ui): prevent infinite refresh loop on user page (fixes #11303)

Python4 giorni fa

12313

BerriAI/litellm38k

fix(responses): forward metadata to Langfuse/callbacks on completion transformation path

Python4 giorni fa

4713

BerriAI/litellm38k

fix(proxy): add TaskRegistry to prevent asyncio task memory leak (fixes #15128)

Python4 giorni fa

35425

BerriAI/litellm38k

feat(proxy): suppress health check access logs via LITELLM_SUPPRESS_HEALTH_LOGS

Python3 giorni fa

225

BerriAI/litellm38k

feat(guardrails): chunk large Bedrock guardrail requests to avoid 429

Python3 giorni fa

57967

BerriAI/litellm38k

fix(proxy): auto-generate Prisma client on startup for custom installs

Python3 giorni fa

46975

BerriAI/litellm38k

fix(copilot): set X-Initiator header correctly for agentic follow-ups

Python3 giorni fa

577348

BerriAI/litellm38k

fix(vertex-ai): remove output_config from Vertex AI Anthropic requests

Python2 giorni fa

20192

BerriAI/litellm38k

fix(mcp): preserve ProxyException status/details in handlers

Pythonun giorno fa

1693

BerriAI/litellm38k

fix(agentcore): avoid premature stop on metadata events

Pythonun giorno fa

1362

BerriAI/litellm38k

fix(pass-through): avoid adapter registry overwrite

Pythonun giorno fa

2,9302,857

BerriAI/litellm38k

fix(cache): preserve embedding indices on cache merges

Pythonun giorno fa

1441

BerriAI/litellm38k

fix(cost_calc): update custom_llm_provider when base_model has different provider prefix

Pythoncirca 8 ore fa

954

BerriAI/litellm38k

fix(responses-bridge): inject reasoning summary when auto_summary enabled without explicit effort

Python4 giorni fa

90

BerriAI/litellm38k

fix(logging): respect turn_off_message_logging for streaming responses

Python4 giorni fa

90

BerriAI/litellm38k

fix(mcp): normalize mounted path server extraction

Pythonun giorno fa

881

BerriAI/litellm38k

perf: use deque for streaming event queues to avoid O(n) pop(0)

Python5 giorni fa

8735

BerriAI/litellm38k

fix(proxy): preserve interactions agent field semantics

Pythonun giorno fa

8561

BerriAI/litellm38k

fix(db): respect DATABASE_SCHEMA env var instead of hardcoding 'public'

Python4 giorni fa

816

BerriAI/litellm38k

fix(a2a): add safe card resolver fallback when a2a-sdk is missing

Python3 giorni fa

6810

BerriAI/litellm38k

fix(ui): parse YAML and update state in blocklist file upload handler

Pythoncirca 8 ore fa

662

BerriAI/litellm38k

fix(github_copilot): check only last message for initiator determination

Pythoncirca 8 ore fa

5812

BerriAI/litellm38k

fix(proxy): preserve master_key set by initialize() during lifespan startup

Python5 giorni fa

541

BerriAI/litellm38k

fix(cost): extract provider from base_model for cross-provider cost calculation

Python5 giorni fa

493

BerriAI/litellm38k

fix(http): display descriptive timeout value instead of 'None seconds'

Python4 giorni fa

467

BerriAI/litellm38k

fix(video): use DEFAULT_VIDEO_ENDPOINT_MODEL in avideo_list provider resolution

Python5 giorni fa

372

BerriAI/litellm38k

fix(proxy): forward retry-after header from upstream providers on 429 responses

Python4 giorni fa

29

BerriAI/litellm38k

fix(auth): remove implicit UI budget cap from CLI JWT

Pythonun giorno fa

292

BerriAI/litellm38k

perf(mcp): replace list.pop(0) with deque.popleft() in MCPEnhancedStreamingIterator

Python5 giorni fa

910

BerriAI/litellm38k

perf(rag): replace list.pop(0) with deque.popleft() in RecursiveCharacterTextSplitter

Python5 giorni fa

32

BerriAI/litellm38k

chore: fix pre-existing ruff lint violations blocking CI

Python2 giorni fa

3

BerriAI/litellm38k

fix(a2a): handle dict messages in context_id propagation

Python2 giorni fa

11

BerriAI/litellm38k

fix(anthropic): strip unsupported schema constraints in tool-based response_format path

Python4 giorni fa

680168

BerriAI/litellm38k

fix(copilot): check only last message for X-Initiator header

Python4 giorni fa

23628

BerriAI/litellm38k

fix: WebSearch interception fails with thinking enabled + SpendLog dedup

Python2 giorni fa

5,78630

BerriAI/litellm38k

fix(bedrock): correct max_output_tokens for moonshotai.kimi-k2.5

Python5 giorni fa

2623

BerriAI/litellm38k

fix(minimax): map reasoning_details to reasoning_content for streaming and non-streaming

Python5 giorni fa

3107

BerriAI/litellm38k

fix(logging): persist provider response headers in StandardLoggingAdditionalHeaders

Python5 giorni fa

949

BerriAI/litellm38k

fix(bedrock): strip anthropic-beta from HTTP headers for Converse API

Python5 giorni fa

36844

BerriAI/litellm38k

fix(anthropic): stop injecting effort-2025-11-24 beta header for output_config on 4.6 models

Python5 giorni fa

5224

BerriAI/litellm38k

feat(responses-bridge): pass thinking_blocks as reasoning items with encrypted_content for OpenAI o-series

Python5 giorni fa

78125

BerriAI/litellm38k

fix(mcp): semantic tool filter matches prefixed tool names

Python5 giorni fa

12825

BerriAI/litellm38k

fix(anthropic): filter unsupported JSON Schema constraints for structured outputs

Python5 giorni fa

71614

BerriAI/litellm37k

Chapter 5 — Unified Context Management: Tests + Documentation

Python5 giorni fa

5144

BerriAI/litellm37k

Chapter 5 — Unified Context Management Across Providers

Python5 giorni fa

4,89519

BerriAI/litellm37k

feat(anthropic): support top-level cache_control for automatic prompt caching

Python5 giorni fa

908

BerriAI/litellm37k

fix(bedrock): inject context_management beta headers in Messages API + tests + docs

Python5 giorni fa

4154

BerriAI/litellm37k

fix: skip effort beta header for Claude 4.6 models

Python5 giorni fa

3812

BerriAI/litellm37k

fix(proxy): apply SERVER_ROOT_PATH to mapped passthrough route matching

Python5 giorni fa

2,9202,826

BerriAI/litellm37k

fix(vertex): preserve 'any type' semantics for JsonValue schemas

Python5 giorni fa

18022

BerriAI/litellm37k

feat(skills): Add OpenAI Skills API support

Python5 giorni fa

4,240178

BerriAI/litellm37k

fix(cost): aimage_edit custom pricing not detected via litellm_metadata

Python5 giorni fa

11312

BerriAI/litellm37k

fix: restrict effort='max' to Claude Opus 4.6 only

Python5 giorni fa

15534

BerriAI/litellm37k

fix: forward reasoning_effort to output_config for Claude 4.6 models

Python5 giorni fa

316146

BerriAI/litellm37k

fix(proxy): catch GuardrailInterventionNormalStringError in apply_guardrail

Python5 giorni fa

4420

stanfordnlp/dspy33k

perf: use collections.deque for history to avoid O(n) pop(0)

Python4 giorni fa

15719

stanfordnlp/dspy33k

feat(adapter): retry on pydantic validation errors (fixes #7693)

Python4 giorni fa

26912

stanfordnlp/dspy33k

feat(adapters): add Video type for video content support

Python3 giorni fa

2733

stanfordnlp/dspy33k

perf(history): replace list with deque for O(1) history updates

Python3 giorni fa

3619

stanfordnlp/dspy33k

feat: retry LM call on pydantic validation and adapter parse errors

Pythoncirca 7 ore fa

294

stanfordnlp/dspy33k

fix: propagate exceptions from async generator in apply_sync_streaming

Pythoncirca 7 ore fa

5

stanfordnlp/dspy32k

fix: propagate exceptions from async generator in apply_sync_streaming

Python5 giorni fa

24

stanfordnlp/dspy32k

fix: `_parse_code` corrupts valid Python code with escape sequences and equals signs

Python5 giorni fa

12630

stanfordnlp/dspy32k

perf: use deque instead of list for history to avoid O(n) pop(0)

Python5 giorni fa

1220

stanfordnlp/dspy32k

feat(pot): add AST validation for early syntax error detection in ProgramOfThought

Python5 giorni fa

71

openai/openai-python30k

feat(responses): add cancel() support for streaming responses

Python3 giorni fa

3242

openai/openai-python30k

fix: handle pydantic ValidationError in beta.chat.completions.parse (fixes #1763)

Python3 giorni fa

1427

openai/openai-python30k

fix(security): redact sensitive headers from debug logs

Python3 giorni fa

9125

openai/openai-python30k

fix(types): make logprobs Optional in response event types (fixes #2489)

Python4 giorni fa

702

openai/openai-python30k

fix(azure): strip model from request body for deployment-based endpoints

Python5 giorni fa

361

microsoft/semantic-kernel27k

Python: fix(python/google): preserve thought_signature in Gemini function call parts

C#5 giorni fa

13714

chroma-core/chroma26k

fix: guard batched_events dict with threading.Lock to prevent KeyError race

Rust5 giorni fa

148

huggingface/smolagents26k

feat(tools): add return_direct parameter to skip LLM processing of tool output

Python4 giorni fa

3705

huggingface/smolagents26k

feat(memory): add automatic memory consolidation via max_memory_steps (fixes #901)

Python4 giorni fa

3705

huggingface/smolagents26k

feat(tools): add transparent async tool support (fixes #334)

Python4 giorni fa

2543

huggingface/smolagents26k

fix(models): guard against missing tool_calls in Bedrock response (fixes #1941)

Python4 giorni fa

831

huggingface/smolagents26k

fix(models): guard against empty choices in all model backends (fixes #1127)

Python4 giorni fa

561

huggingface/smolagents26k

fix: handle consecutive system messages with plain string content

Python5 giorni fa

482

huggingface/smolagents26k

fix: reset _print_outputs before ast.parse to prevent SyntaxError print leak

Python5 giorni fa

217

modelcontextprotocol/python-sdk22k

fix(oauth): include client_id in token request body for client_secret_post

Python5 giorni fa

692

modelcontextprotocol/python-sdk22k

fix: collapse single-error ExceptionGroups from task group cancellations

Python5 giorni fa

2168

modelcontextprotocol/python-sdk22k

fix: catch ClosedResourceError when sending error log to disconnected client

Python5 giorni fa

575

PrefectHQ/prefect22k

fix: correct return type annotation for dbt_runner_freshness_success fixture

Python5 giorni fa

11

google/adk-python18k

fix: handle Pydantic BaseModel in _safe_json_serialize for telemetry

Pythoncirca 7 ore fa

44

camel-ai/camel16k

fix(screenshot): avoid re-entrant agent.step() in read_image

Python5 giorni fa

36242

camel-ai/camel16k

fix: generalize rate-limit retry to all LLM providers

Python5 giorni fa

31380

pydantic/pydantic-ai15k

fix(openai): remove minLength/maxLength from strict-incompatible keys

Python4 giorni fa

313

pydantic/pydantic-ai15k

fix: handle malformed tool call args in Anthropic retry path

Python5 giorni fa

549325

pydantic/pydantic-ai15k

fix: standardize ToolCallPartDelta.part_delta_kind to 'tool-call'

Python5 giorni fa

644472

pydantic/pydantic-ai15k

fix: set run_id on ModelRequest when resuming without prompt

Python5 giorni fa

452

pydantic/pydantic-ai15k

fix: LogfirePlugin compat with temporalio >= 1.23.0 (interceptors renamed)

Python5 giorni fa

274

567-labs/instructor12k

fix(security): prevent prompt injection in llm_validator via XML delimiters

Python5 giorni fa

831

567-labs/instructor12k

fix(bedrock): pass through cachePoint content blocks for prompt caching

Python5 giorni fa

3211

wandb/wandb11k

perf: use deque instead of list for _pending_jobs in StepUpload

Pythoncirca 8 ore fa

43

llamastack/llama-stack8.3k

fix: guard against NoneType in content join for tool-call messages

Python5 giorni fa

522

github/copilot-sdk7.6k

fix(python): correct PermissionHandler.approve_all type annotations

TypeScript5 giorni fa

42

strands-agents/sdk-python5.2k

fix(mcp): add timeout to background thread join in stop() to prevent hangs

Python5 giorni fa

445

strands-agents/sdk-python5.2k

fix: skip AgentState reset for MultiAgentBase executors in GraphNode

Python5 giorni fa

552

strands-agents/sdk-python5.2k

fix: suppress PydanticSerializationUnexpectedValue warnings for Anthropic streaming

Python5 giorni fa

44

anthropics/anthropic-sdk-python2.9k

perf: cache type introspection in _transform_recursive to eliminate redundant dispatch

Python4 giorni fa

26079

BerriAI/litellm

fix: exclude gpt-5.2-chat from temperature passthrough

Python6 giorni fa

1,061179

BerriAI/litellm

fix(anthropic): handle OAuth tokens in count_tokens endpoint

Python6 giorni fa

5,666854

BerriAI/litellm

fix(azure): forward realtime_protocol from config and relax api_version check for GA path

Python6 giorni fa

5,692855

BerriAI/litellm

fix(anthropic): populate output_config when reasoning_effort is used on Claude 4.6

Python5 giorni fa

265

openai/openai-agents-python

perf: use deque for BFS queue and voice task ordering

Python6 giorni fa

86

BerriAI/litellm

fix: defer remote model cost map fetch to first use (simplified)

Python5 giorni fa

21715

strands-agents/sdk-python

fix: support Bedrock application inference profile ARNs in cache detection

Python5 giorni fa

1742

pydantic/pydantic-ai

fix: detect empty items schema as strict-incompatible in OpenAI transformer

Python5 giorni fa

521324

pydantic/pydantic-ai

fix: close async generator in run_stream_sync for OTel span finalization

Python5 giorni fa

1337

livekit/agents

fix: recover from truncated JSON in tool call arguments

Python5 giorni fa

1291

pydantic/pydantic-ai

fix: add missing strict-incompatible JSON Schema keys for OpenAI

Python5 giorni fa

1205

langchain-ai/langchain

fix: preserve tool_use/tool_result pairs in trim_messages

Python5 giorni fa

1802

modelcontextprotocol/python-sdk

fix: collapse BaseExceptionGroup to surface real errors from task groups

Python5 giorni fa

26522

dottxt-ai/outlines

fix(dsl): add JSON double-quotes to string literals inside container types

Python5 giorni fa

2331,992

BerriAI/litellm

fix: extract router_model_id from metadata for image cost calc

Python6 giorni fa

173

BerriAI/litellm

fix: defer remote model cost map HTTP fetch from import time to first access

Python6 giorni fa

4602

modelcontextprotocol/typescript-sdk

fix: accumulate OAuth scopes across 401/403 for progressive authorization

TypeScript6 giorni fa

992

pydantic/pydantic-ai

fix: handle malformed JSON tool args in Anthropic retry path

Python6 giorni fa

537329

vercel/ai

fix(gateway): normalize V2 flat usage to V3 nested format

TypeScript6 giorni fa

165

pydantic/pydantic-ai

fix: add missing strict-incompatible JSON Schema keys to OpenAIJsonSchemaTransformer

Python6 giorni fa

622330

modelcontextprotocol/python-sdk

fix: handle tools/resources/prompts list_changed notifications in client

Python6 giorni fa

2241

vercel/ai

fix(anthropic): convert oneOf to anyOf in JSON schemas for Anthropic API

TypeScript6 giorni fa

2073

pydantic/pydantic-ai

fix: wrap Google streaming errors as ModelHTTPError/ModelAPIError

Python6 giorni fa

1592

BerriAI/litellm

fix(moonshot): validate reasoning_content in multi-turn tool calling

Python6 giorni fa

201

pydantic/pydantic-ai

fix(xai): promote cache_read_tokens to top-level RequestUsage field

Python6 giorni fa

526338

pydantic/pydantic-ai

fix: align ToolCallPartDelta.part_delta_kind with ToolCallPart.part_kind

Python6 giorni fa

611471

agno-agi/agno

fix: prevent TypeError when MCP tool parameters collide with framework-injected names

Python6 giorni fa

1243

vercel/ai

fix(ai): throw AbortError from generateText() instead of silently swallowing it

TypeScript6 giorni fa

2057

vercel/ai

fix(ai): guard toolResultsStreamController against enqueue/close on closed controller

TypeScript6 giorni fa

30917

vercel/ai

fix(anthropic): validate thinking block requirement and prevent silent reasoning drop

TypeScript6 giorni fa

21017

langchain-ai/langchainjs

fix: throw descriptive error when Google AI returns empty generations

TypeScript6 giorni fa

1119

modelcontextprotocol/typescript-sdk

fix: handle URL-encoded token responses from OAuth providers

TypeScript6 giorni fa

991

vercel/ai

fix(openai): properly handle stream errors in Responses API

TypeScript6 giorni fa

13513

vercel/ai

fix(ui): preserve partial message and metadata on stream error

TypeScript6 giorni fa

102

vercel/ai

fix(openai-compatible): support audio URLs, non-wav/mp3 audio, and video media types

TypeScript6 giorni fa

14546

modelcontextprotocol/typescript-sdk

fix(server): auto-unwrap z.object() schemas in server.tool()

TypeScript6 giorni fa

1464

vercel/ai

fix(react): prevent stale transport body in useChat

TypeScript6 giorni fa

951

langchain-ai/langchain

fix(core): fix RunnableRetry.batch position corruption with return_exceptions

Python5 giorni fa

9414

vercel/ai

fix(telemetry): add missing usage attributes to generateText and generateObject

TypeScript6 giorni fa

914

modelcontextprotocol/typescript-sdk

fix: throw error when plain JSON Schema is passed as inputSchema to server.tool()

TypeScript6 giorni fa

91

vercel/ai

fix: add missing usage telemetry attributes to generateText and generateObject

TypeScript6 giorni fa

8619

guardrails-ai/guardrails-api

fix(openai): include id, object, created, model in chat completions response

Python5 giorni fa

84

langchain-ai/langchain

fix(core): RunnableRetry.batch corrupted outputs on partial retry success

Python6 giorni fa

842

modelcontextprotocol/typescript-sdk

fix: accumulate OAuth scopes across 401/403 responses for progressive authorization

TypeScript6 giorni fa

832

modelcontextprotocol/python-sdk

fix: accumulate OAuth scopes across 401/403 for progressive authorization

Python6 giorni fa

7711

CopilotKit/CopilotKit

fix: guard against non-object tool arguments from LLMs

TypeScript6 giorni fa

755

BerriAI/litellm

fix: respect disable_exception_on_block in apply_guardrail for Bedrock guardrails

Python6 giorni fa

7219

vercel/ai

fix(generate-image): aggregate numeric gateway metadata across multi-call splits

TypeScript6 giorni fa

724

strands-agents/sdk-python

fix: preserve reasoning content signature from LiteLLM thinking models

Python5 giorni fa

70

vercel/ai

fix(openai): accept empty string type in streaming tool call deltas (Azure)

TypeScript6 giorni fa

691

agno-agi/agno

fix: prevent race condition in MCPTools session creation with header_provider

Python5 giorni fa

6754

vercel/ai

fix(amazon-bedrock): support cross-region inference prefixes for embedding models

TypeScript6 giorni fa

672

modelcontextprotocol/python-sdk

fix: catch ClosedResourceError in _handle_message error recovery path

Python6 giorni fa

665

modelcontextprotocol/typescript-sdk

fix: allow dynamic handler registration after connect with pre-supplied capabilities

TypeScript6 giorni fa

661

vercel/ai

fix(processUIMessageStream): pass through providerMetadata on file chunks

TypeScript6 giorni fa

651

pydantic/pydantic-ai

fix: detect empty `items: {}` as strict-incompatible in OpenAIJsonSchemaTransformer

Python6 giorni fa

64

BerriAI/litellm

fix(openai): drop temperature for gpt-5.2-chat variants with drop_params

Python5 giorni fa

626

pydantic/pydantic-ai

fix: match dict values in SpanQuery has_attributes

Python5 giorni fa

611

modelcontextprotocol/typescript-sdk

fix: registerCapabilities registers logging/setLevel handler

TypeScript6 giorni fa

6110

vercel/ai

fix: emit ai.toolCall telemetry span for tools without execute

TypeScript6 giorni fa

584

BerriAI/litellm

fix(anthropic): remove spurious effort beta header for output_config and restrict effort=max to Opus 4.6

Python6 giorni fa

5618

stanfordnlp/dspy

fix: prevent _parse_code from corrupting valid Python code

Python6 giorni fa

5511

pydantic/pydantic-ai

fix: detect empty `items: {}` as strict-incompatible for arrays

Python5 giorni fa

54

vercel/ai

fix(provider-utils): RFC 9110 compliant User-Agent header format

TypeScript6 giorni fa

538

crewAIInc/crewAI

fix(task): parse pydantic output before guardrail invocation

Python5 giorni fa

5110

langchain-ai/langchain

fix(openai): ensure stream=False in payload when disable_streaming is active

Python5 giorni fa

51

pydantic/pydantic-ai

fix: SpanQuery has_attributes matches dict values stored as JSON strings

Python6 giorni fa

501

huggingface/smolagents

fix: strip incomplete closing tag before appending in CodeAgent streaming

Python6 giorni fa

491

openai/openai-python

fix: handle ValidationError in structured output parsing gracefully

Python5 giorni fa

471

vercel/ai

fix(google): infer type from enum values when not explicitly set

TypeScript6 giorni fa

47

microsoft/semantic-kernel

Python: fix: ChatHistoryTruncationReducer orphans TOOL role messages

C#5 giorni fa

461

pydantic/pydantic-ai

fix: assign run_id to ModelRequest when resuming without prompt

Python6 giorni fa

452

crewAIInc/crewAI

fix(telemetry): skip signal handler registration in non-main threads

Python5 giorni fa

44

pydantic/pydantic-ai

fix(evals): HasMatchingSpan YAML roundtrip fails when single field is a BaseModel

Python6 giorni fa

445

modelcontextprotocol/python-sdk

fix: reject JSON-RPC requests with id: null instead of misclassifying as notification

Python6 giorni fa

431

BerriAI/litellm

fix: update custom_llm_provider when base_model provider differs from deployment

Python6 giorni fa

42

modelcontextprotocol/typescript-sdk

fix: add missing size field to ResourceSchema per MCP spec

TypeScript6 giorni fa

42

CopilotKit/CopilotKit

fix(react): replace static KaTeX CSS import with dynamic runtime injection

TypeScript6 giorni fa

401

stanfordnlp/dspy

fix: handle non-AnyIO threads in sync_send_to_stream for dspy.Parallel

Python5 giorni fa

385

langchain-ai/langgraph

fix: add warning logs when JsonPlusSerializer silently returns None

Python6 giorni fa

38

getzep/graphiti

fix(search): wrap group_ids OR clause in parentheses for BM25 queries

Python6 giorni fa

383

stanfordnlp/dspy

fix: share usage_tracker across parallel workers for correct aggregation

Python5 giorni fa

3722

stanfordnlp/dspy

fix: propagate exceptions from async generator in apply_sync_streaming

Python6 giorni fa

34

modelcontextprotocol/python-sdk

fix: gracefully terminate active sessions on shutdown

Python5 giorni fa

31

huggingface/smolagents

fix: propagate contextvars through timeout decorator for OTel span parenting

Python5 giorni fa

311

BerriAI/litellm

fix(video): use DEFAULT_VIDEO_ENDPOINT_MODEL in avideo_list provider resolution

Python6 giorni fa

301

ag-ui-protocol/ag-ui

fix(mastra): handle non-JSON working memory templates gracefully

Python6 giorni fa

2511

stanfordnlp/dspy

fix: support TRL >= 0.16 by using max_length instead of deprecated max_seq_length

Python5 giorni fa

245

modelcontextprotocol/python-sdk

fix: detect Context parameter in callable class instances

Python6 giorni fa

241

Chainlit/chainlit

fix(data): prevent empty string from overwriting step content on upsert

Python5 giorni fa

232

livekit/agents

fix(anthropic): strip trailing assistant messages for Claude 4.6+

Python5 giorni fa

232

vercel/ai

fix(langchain): HITL interrupt uses original tool call ID from messages stream

TypeScript6 giorni fa

2117

langchain-ai/langchain

fix(core): guard merge_lists against non-dict elements when matching by index

Python6 giorni fa

181

BerriAI/litellm

perf: use deque for streaming event queues to avoid O(n) pop(0)

Python6 giorni fa

1716

pydantic/pydantic-ai

fix: use global retry count for output validators on tool output path

Python6 giorni fa

173

567-labs/instructor

fix: make diskcache an optional dependency (CVE-2025-69872)

Python6 giorni fa

1616

apache/airflow

perf: use deque for cursor row buffers in Google provider

Python6 giorni fa

1613

dottxt-ai/outlines

fix(llamacpp): correct EOS attention mask and vocab truncation bugs

Python5 giorni fa

156

vercel/ai

fix(google): make urlMetadata optional in urlContextMetadata schema

TypeScript6 giorni fa

156

modelcontextprotocol/python-sdk

feat(server): expose progress_callback on ServerSession request methods

Python6 giorni fa

141

chroma-core/chroma

fix: guard batched_events dict with threading.Lock to prevent KeyError race

Rust6 giorni fa

148

camel-ai/camel

perf: use deque for BFS traversal and tool buffer processing

Python6 giorni fa

1412

PostHog/posthog

perf: use deque for FIFO queues across batch exports, signals, CDP, and lineage

Python6 giorni fa

1414

ray-project/ray

perf: use deque for BFS queues and FIFO buffers across core modules

Python6 giorni fa

1411

huggingface/smolagents

fix: handle consecutive system messages with string content in get_clean_message_list

Python6 giorni fa

143

docling-project/docling

fix(readingorder): assign FURNITURE content_layer to footer/header in container groups

Python5 giorni fa

131

huggingface/transformers

fix: suppress ANSI escape codes when stdout is not a terminal

Python6 giorni fa

136

openai/openai-python

fix: move unreachable error event check before thread event guard in streaming

Python6 giorni fa

128

huggingface/datasets

perf: use deque for async map task/index queues

Python6 giorni fa

1111

modelcontextprotocol/python-sdk

fix: preserve HTTP status code in streamable HTTP client error messages

Python6 giorni fa

101

vercel/ai

fix(provider-utils): include type: 'object' in default tool schema

TypeScript6 giorni fa

101

weaviate/weaviate-python-client

perf: use deque for iterator object cache

Python6 giorni fa

98

deepspeedai/DeepSpeed

perf: use deque for FIFO queues in sequence parallel, superoffload, and compile

Python6 giorni fa

98

ag-ui-protocol/ag-ui

fix(client): guard process.env access for browser compatibility

Python6 giorni fa

92

modelcontextprotocol/python-sdk

fix(stdio): allow configurable memory-stream buffer size

Python6 giorni fa

83

BerriAI/litellm

perf(mcp): replace list.pop(0) with deque.popleft() in MCPEnhancedStreamingIterator

Python6 giorni fa

87

gradio-app/gradio

perf: use deque for SSE pending message queues in gradio_client

Python6 giorni fa

86

modelcontextprotocol/python-sdk

fix: handle ClosedResourceError when logging errors to disconnected clients

Python6 giorni fa

85

deepset-ai/haystack

perf: use deque for sentence span merging in SentenceSplitter

MDX6 giorni fa

75

Significant-Gravitas/AutoGPT

perf: use deque for FIFO tool-output and execution-output queues

Python6 giorni fa

76

modelcontextprotocol/python-sdk

fix: dup file descriptors in stdio_server to avoid closing real stdin/stdout

Python6 giorni fa

76

crewAIInc/crewAI

fix: initialize tools list when None before extending with MCP/platform tools

Python6 giorni fa

62

PrefectHQ/prefect

perf: use deque for BFS queues in dbt freshness and settings scripts

Python6 giorni fa

64

sympy/sympy

perf: use deque for AndOr_Base.flatten queue

Python6 giorni fa

66

run-llama/llama_index

fix(google-genai): replace asyncio.run() with asyncio_run() to support nested event loops

Python6 giorni fa

65

modelcontextprotocol/python-sdk

perf: use deque for InMemoryTaskMessageQueue FIFO operations

Python6 giorni fa

54

langchain-ai/langchain

perf(text-splitters): use deque for line consumption in markdown splitter

Python6 giorni fa

54

langchain-ai/langchain

perf(core): use deque for result reassembly in RunnableRetry batch

Python6 giorni fa

52

qdrant/qdrant-client

perf: replace list.pop(0) with deque.popleft() in ModelEmbedder

Python6 giorni fa

55

vercel/ai

fix(openai): omit reasoning item id when store is false

TypeScript6 giorni fa

514

getzep/graphiti

fix(falkordb): use startNode/endNode instead of re-MATCH for edge search

Python5 giorni fa

44

567-labs/instructor

fix(bedrock): pass through cachePoint dicts in content items

Python6 giorni fa

4

docling-project/docling

fix(vlm): convert images to RGB before passing to transformers

Python5 giorni fa

4

langchain-ai/langchain

perf(core): use deque for pipeline batch output reassembly in RunnableSequence

Python6 giorni fa

44

guidance-ai/guidance

perf: use deque for BFS/DFS traversal in TraceNode.traverse()

Jupyter Notebook6 giorni fa

43

modal-labs/modal-client

perf: use deque for _watch_output_buffer in FileIO

Python6 giorni fa

43

pytorch/pytorch

perf: use deque for activation offloading reload queue

Python6 giorni fa

43

microsoft/semantic-kernel

Python: fix(google-ai): skip api_key check when use_vertexai is True

C#5 giorni fa

33

modelcontextprotocol/python-sdk

fix: use os.dup() to prevent stdio transport from closing real stdin/stdout

Python6 giorni fa

32

BerriAI/litellm

perf(rag): replace list.pop(0) with deque.popleft() in RecursiveCharacterTextSplitter

Python6 giorni fa

32

HKUDS/LightRAG

perf: use deque for BFS queue in get_knowledge_subgraph()

Python6 giorni fa

32

FoundationAgents/MetaGPT

perf: use deque for BFS queue in get_function_dependency()

Python6 giorni fa

32

simonw/llm

perf: use deque for async response chunk iteration

Python6 giorni fa

32

wandb/wandb

perf: use deque for pending upload job queue in StepUpload

Python6 giorni fa

32

marimo-team/marimo

perf: use deque for BFS queue in pytest_changed dependency traversal

Python6 giorni fa

32

apache/hamilton

perf: use deque for BFS queue in graph traversal

Jupyter Notebook6 giorni fa

32

scrapy/scrapy

perf: use deque for waiting_deferreds FIFO queue in DeferredAsyncIterator

Python6 giorni fa

32

aio-libs/aiohttp

perf: use deque for FormData.add_fields() queue

Python6 giorni fa

32

ag-ui-protocol/ag-ui

fix(python): align ReasoningMessageStartEvent.role with TypeScript SDK

Python6 giorni fa

22

openai/openai-python

fix: bound TypeAdapter lru_cache to prevent memory leak in multi-threaded usage

Python5 giorni fa

11

huggingface/smolagents

fix: from_dict drops child agent authorized_imports

Python6 giorni fa

11

huggingface/smolagents

fix(bedrock): use .get() for tool_calls to prevent KeyError on text-only responses

Python6 giorni fa

11

567-labs/instructor

fix: make diskcache an optional dependency

Python6 giorni fa

11

ag-ui-protocol/ag-ui

fix(python): make Tool.parameters optional to match TypeScript SDK

Python6 giorni fa

11

vercel/ai

fix(openai): omit reasoning item id when store is false

TypeScript6 giorni fa

14

pydantic/pydantic-ai

fix: standardize ToolCallPartDelta.part_delta_kind to 'tool-call'

Python5 giorni fa

611471

modelcontextprotocol/python-sdk

feat(client): handle list_changed notifications via callbacks

Python5 giorni fa

3702

run-llama/llama_index

fix: handle string output in StructuredLLM.chat and achat

Python5 giorni fa

23135

google/adk-python

fix: stabilize function call IDs across streaming events

Python6 giorni fa

2326

BerriAI/litellm

fix(vertex): preserve empty schemas for JsonValue/Any types in Gemini

Python5 giorni fa

2577

BerriAI/litellm

fix(minimax): map reasoning_details to reasoning_content for streaming and non-streaming

Python5 giorni fa

1862

google/adk-python

fix: preserve thought_signature in LiteLLM function call round-trip

Python5 giorni fa

952

camel-ai/camel

fix(screenshot_toolkit): prevent re-entrant agent.step() in read_image

Python5 giorni fa

932

camel-ai/camel

fix(agents): generalize RateLimitError to catch non-OpenAI rate limits

Python6 giorni fa

892

langchain-ai/langchain

fix(core): prevent RunnableRetry.batch from returning corrupted outputs on partial failure

Python6 giorni fa

84

mlflow/mlflow

fix(tracing): include auth credentials in OTLP exporter headers

Python6 giorni fa

845

huggingface/smolagents

fix: extract toolUse blocks from Bedrock Converse API response

Python6 giorni fa

744

pydantic/pydantic-ai

fix: handle malformed tool args in Anthropic message reconstruction

Python5 giorni fa

705

langchain-ai/langchain

fix(core): cancel pending tasks in abatch_as_completed on exception or early exit

Python6 giorni fa

642

huggingface/smolagents

fix: handle string content when merging consecutive same-role messages

Python5 giorni fa

6410

CooperAgent/cooper

feat: add /delegate command for cloud agent task delegation (#19)

TypeScript19 giorni fa

62

pydantic/pydantic-ai

fix(evals): SpanQuery has_attributes matches dict/list values against JSON-serialized form

Python6 giorni fa

581

modelcontextprotocol/python-sdk

fix: handle ClosedResourceError when logging errors to disconnected clients

Python6 giorni fa

545

modelcontextprotocol/python-sdk

fix: preserve HTTP status codes in streamable HTTP transport error responses

Python6 giorni fa

5038

modelcontextprotocol/python-sdk

fix: detect Context parameter in callable class instances

Python5 giorni fa

501

strands-agents/sdk-python

fix(anthropic): avoid Pydantic serialization warnings for ParsedTextBlock in message_stop events

Python5 giorni fa

494

BerriAI/litellm

fix(proxy): apply SERVER_ROOT_PATH to mapped passthrough route matching

Python6 giorni fa

481

huggingface/smolagents

fix: handle string content in consecutive same-role message merging

Python5 giorni fa

464

browser-use/browser-use

fix: make tunnel process management Windows-compatible

Python5 giorni fa

4517

agno-agi/agno

fix(chromadb): filter empty documents and ensure unique IDs in insert/upsert

Python6 giorni fa

4018

strands-agents/sdk-python

fix(graph): prevent reset_executor_state from corrupting MultiAgentBase state

Python5 giorni fa

391

BerriAI/litellm

fix(cost): extract provider from base_model for cross-provider cost calculation

Python5 giorni fa

381

pydantic/pydantic-ai

fix: add missing strict-incompatible JSON Schema keys to OpenAI transformer

Python5 giorni fa

34

huggingface/smolagents

fix: handle consecutive messages with string content in get_clean_message_list

Python5 giorni fa

3410

BerriAI/litellm

fix(anthropic): restrict effort='max' to Opus 4.6 only

Python6 giorni fa

343

BerriAI/litellm

fix: update custom_llm_provider when base_model has a different provider

Python6 giorni fa

34

BerriAI/litellm

fix(anthropic): restrict effort='max' to Opus 4.6 only

Python5 giorni fa

343

strands-agents/sdk-python

fix(mcp): add timeout to background thread join in stop() to prevent hangs

Python5 giorni fa

315

BerriAI/litellm

fix(anthropic): restrict effort='max' to Opus 4.6 only

Python5 giorni fa

293

stanfordnlp/dspy

fix: propagate exceptions from apply_sync_streaming instead of swallowing

Python5 giorni fa

27

huggingface/smolagents

fix: reset PrintContainer before ast.parse to prevent output leaking across steps

Python5 giorni fa

257

huggingface/smolagents

fix: reset PrintContainer before ast.parse to prevent output leak on SyntaxError

Python6 giorni fa

237

stanfordnlp/dspy

fix: propagate exceptions from async generator in apply_sync_streaming

Python5 giorni fa

22

run-llama/llama_index

fix(openai-like): strip chat-only kwargs in completion fallback path

Python6 giorni fa

224

huggingface/smolagents

fix: reset print outputs before parsing to prevent leak on SyntaxError

Python5 giorni fa

217

stanfordnlp/dspy

fix: replace list.pop(0) with deque.popleft() for O(1) history eviction

Python5 giorni fa

2120

huggingface/smolagents

fix: reset PrintContainer before ast.parse to prevent output leak on SyntaxError

Python6 giorni fa

217

huggingface/smolagents

fix: reset PrintContainer before ast.parse to prevent output leak on SyntaxError

Python6 giorni fa

217

huggingface/smolagents

fix: handle None content in parse_code_blobs and extract_code_from_text

Python6 giorni fa

202

BerriAI/litellm

fix(ui): parse and save blocked words from uploaded YAML file

Python6 giorni fa

181

BerriAI/litellm

fix(anthropic): remove spurious effort-2025-11-24 beta header for Claude 4.6 models

Python5 giorni fa

1716

BerriAI/litellm

fix(anthropic): set output_config.effort when reasoning_effort maps to adaptive for 4.6 models

Python5 giorni fa

17

BerriAI/litellm

fix(ui): parse YAML blocklist file and populate blocked words

Python5 giorni fa

161

google/adk-python

fix(sessions): skip unnecessary FOR UPDATE lock on app/user state rows

Python5 giorni fa

167

browser-use/browser-use

fix: support RFC 6266 filename* in Content-Disposition for non-ASCII filenames

Python5 giorni fa

144

stanfordnlp/dspy

perf: use deque instead of list for LM history to avoid O(n) pop(0)

Python5 giorni fa

1313

run-llama/llama_index

fix(openai): omit tool params from request when no tools provided

Python5 giorni fa

119

mem0ai/mem0

fix: only pass dimensions to OpenAI embeddings API when explicitly configured

Python6 giorni fa

119

run-llama/llama_index

fix(openai): omit parallel_tool_calls when no tools are provided

Python5 giorni fa

109

google/adk-python

fix(cli): filter non-agent directories from list_agents

Python6 giorni fa

10

run-llama/llama_index

fix: omit parallel_tool_calls when no tools to prevent null in API request

Python6 giorni fa

99

mem0ai/mem0

fix: _safe_deepcopy_config no longer strips http_auth from OpenSearch configs

Python5 giorni fa

82

run-llama/llama_index

fix(openai): strip chat-only params from completions endpoint

Python6 giorni fa

8

BerriAI/litellm

fix(proxy): preserve master_key set by initialize() during lifespan startup

Python5 giorni fa

71

BerriAI/litellm

fix(anthropic): populate output_config when reasoning_effort is used with Claude 4.6

Python6 giorni fa

71

browser-use/browser-use

fix: use tempfile.gettempdir() instead of hardcoded /tmp

Python5 giorni fa

64

agno-agi/agno

fix(litellm): prevent empty strings from overwriting tool metadata in streaming

Python6 giorni fa

66

google/adk-python

fix(telemetry): serialize Pydantic BaseModel in _safe_json_serialize

Python6 giorni fa

63

browser-use/browser-use

fix: use tempfile.gettempdir() instead of hardcoded /tmp for downloads path

Python5 giorni fa

64

agno-agi/agno

fix: avoid keyword argument collision in MCP tool entrypoints

Python6 giorni fa

53

langchain-ai/langchain

fix(openai): strip stream param in _generate/_agenerate to fix disable_streaming crash

Python6 giorni fa

42

browser-use/browser-use

fix: use tempfile.gettempdir() instead of hardcoded /tmp in BrowserProfile

Python6 giorni fa

43

huggingface/smolagents

fix: clear print outputs before syntax check to prevent stale output leaking

Python5 giorni fa

43

Textualize/textual

perf: use deque for lazy widget reveal queue

Python6 giorni fa

43

agno-agi/agno

fix: prevent empty string from overwriting tool name in streaming parse_tool_calls

Python6 giorni fa

33

google/adk-python

fix: remove developer note from transfer_to_agent tool description

Python5 giorni fa

35

BerriAI/litellm

fix(ui): preserve newlines in tool descriptions on logs page

Python5 giorni fa

22

agno-agi/agno

fix(embedder): wrap AzureOpenAIEmbedder input in list for non-default models

Python6 giorni fa

22

BerriAI/litellm

fix(ui): render newlines in tool descriptions in log viewer

Python6 giorni fa

22

run-llama/llama_index

fix(openai-like): default is_chat_model to True for modern API compatibility

Python5 giorni fa

22

agno-agi/agno

fix: correct SiliconFlow base_url from .com to .cn

Python6 giorni fa

11

mem0ai/mem0

fix: wrap vector and payload in lists when calling insert from update

Python6 giorni fa

11

browser-use/browser-use

fix: filter chrome-extension:// targets from get_all_page_targets

Python6 giorni fa

11

agno-agi/agno

fix: correct SiliconFlow default base_url from .com to .cn

Python5 giorni fa

11

Dashboard

Pull Requests Recenti

fix(core): prevent RunnableRetry.batch output corruption when partial failures occur

fix(openai): prevent crash when disable_streaming="tool_calling" in streaming mode

fix(core): pass bound tools to token counter in trim_messages

fix(core): make trim_messages respect tool-call group integrity

fix(config): RAG_OPENAI env vars now correctly inherit OPENAI_API values

fix(config): RAG_OPENAI env vars now correctly inherit OPENAI_API values

fix: use tempfile.gettempdir() instead of hardcoded /tmp for Windows compat

fix(openai): exclude parallel_tool_calls from request when no tools provided

fix(flow): allow Flow subclasses in crew.copy() for kickoff_for_each (fixes #4555)

fix(crew): preserve full task output in crew.kickoff() (fixes #4603)

fix(cli): memory leak in Repository.is_git_repo() due to @lru_cache

fix: wrap input as list for non-default Azure models

fix: prevent parameter name collision between framework and MCP tool args

fix: parse raw JSON before cleaning to preserve fenced code blocks (fixes #5901)

fix: add max_history_tokens to prevent session overflow (fixes #5011)

fix: resolve output_schema and reasoning conflict for Anthropic (fixes #5582)

fix: prevent deadlock in parallel MCP tool calls with header_provider (fixes #6094)

fix(litellm): prevent empty tool name from overwriting valid name in streaming

fix(tools): gracefully repair truncated JSON in tool call arguments

fix: WebSearch interception fails with thinking enabled + SpendLog dedup

fix(streaming): prevent Vertex AI Claude content truncation when finish_reason races content

fix(featherless_ai): use correct FEATHERLESS_AI_API_KEY env var name

fix(ui): parse YAML blocklist file and populate blocked words

fix(proxy): bound spend_log_transactions to prevent OOM (#21514)

fix(cost): fix Vercel AI Gateway cost tracking returning $0

fix(adapter): clamp max_tokens on Anthropic pass-through to backend model limit

fix(router): resolve Gemini CLI model variant suffixes (-customtools)

fix(cost): audio transcription cost always 0 for custom-priced models

fix(proxy): add default max_idle_connection_lifetime to prevent stale PostgreSQL connections

fix(proxy): persist default budget_id for implicitly created end users

fix(anthropic-passthrough): defer first content_block_start to respect thinking block ordering

fix(router): downgrade Azure model identification from ERROR to DEBUG

fix(http): add litellm.aclose() and sync connector cleanup for session leaks

fix(logging): prevent RuntimeError when event loop changes across tests

fix(alerting): prevent false 'Requests are hanging' Slack alerts

fix(proxy): strip null bytes from spend log data before PostgreSQL insertion

fix(streaming): propagate provider cost through streaming chunk builder

fix(proxy): remove wildcard patterns from /models endpoint response

fix(bedrock): normalize double-wrapped embedding input to prevent JSONArray error

fix(streaming): thread-safe model_copy to prevent dictionary changed size error

fix(db): fix DailyTagSpend view double-counting with multi-tag requests

fix(cost): fix rerank cost calculation always returning 0

fix(logging): enqueue async callbacks synchronously to prevent task cancellation (#8842)

fix: merge x-litellm-tags with team/project tags instead of overwriting (fixes #14052)

fix: preserve 400 status code for context window errors in streaming (fixes #18689)

fix(logging): stop configuring root/external loggers (fixes #16284)

fix(proxy): prevent OOM on Virtual Keys UI with large key counts (#19477)

fix(http): add connection pool limits to prevent file descriptor exhaustion (fixes #13220)

fix(otel): decouple OTel tracing from proxy dependencies (fixes #13081)

fix(ui): prevent infinite refresh loop on user page (fixes #11303)

fix(responses): forward metadata to Langfuse/callbacks on completion transformation path

fix(proxy): add TaskRegistry to prevent asyncio task memory leak (fixes #15128)

feat(proxy): suppress health check access logs via LITELLM_SUPPRESS_HEALTH_LOGS

feat(guardrails): chunk large Bedrock guardrail requests to avoid 429

fix(proxy): auto-generate Prisma client on startup for custom installs

fix(copilot): set X-Initiator header correctly for agentic follow-ups

fix(vertex-ai): remove output_config from Vertex AI Anthropic requests

fix(mcp): preserve ProxyException status/details in handlers

fix(agentcore): avoid premature stop on metadata events

fix(pass-through): avoid adapter registry overwrite

fix(cache): preserve embedding indices on cache merges

fix(cost_calc): update custom_llm_provider when base_model has different provider prefix

fix(responses-bridge): inject reasoning summary when auto_summary enabled without explicit effort

fix(logging): respect turn_off_message_logging for streaming responses

fix(mcp): normalize mounted path server extraction

perf: use deque for streaming event queues to avoid O(n) pop(0)

fix(proxy): preserve interactions agent field semantics

fix(db): respect DATABASE_SCHEMA env var instead of hardcoding 'public'

fix(a2a): add safe card resolver fallback when a2a-sdk is missing

fix(ui): parse YAML and update state in blocklist file upload handler

fix(github_copilot): check only last message for initiator determination

fix(proxy): preserve master_key set by initialize() during lifespan startup

fix(cost): extract provider from base_model for cross-provider cost calculation

fix(http): display descriptive timeout value instead of 'None seconds'

fix(video): use DEFAULT_VIDEO_ENDPOINT_MODEL in avideo_list provider resolution

fix(proxy): forward retry-after header from upstream providers on 429 responses

fix(auth): remove implicit UI budget cap from CLI JWT

perf(mcp): replace list.pop(0) with deque.popleft() in MCPEnhancedStreamingIterator