v0.61.0 (2026-05-31)

v0.61.0 (2026-05-31)

Feat

• sdk: expose use_legacy_attributes via Traceloop.init() (#4133)
• bedrock: add async support for aioboto3 (#4135)
• openai-agents: GenAI semconv compliance (#3837)

Fix

• openai: instrument responses.parse() for structured-output tracing (#4198)
• langchain,anthropic,groq,mistralai,bedrock,ollama,sagemaker,together: record exceptions and set ERROR status on failed spans (#4101)
• vector-db: emit consistent embeddings_count, result_count, similarity attrs across Pinecone and Milvus (#1870) (#4156)
• sdk: warn when both exporter and processor are passed to Traceloop.init() (#4137)
• openai-agents: capture response.instructions as system prompt in generation spans (#4131)
• openai-agents: emit cache_read.input_tokens and reasoning_tokens (#4130)
• chromadb: emit one result event per document across all queries (#4105)
• mcp: set error.type attribute on protocol-level tool errors (#4103)
• sdk: replace deprecated Pydantic .json() and double-encoding .model_dump_json() in JSONEncoders (#4098)
• chromadb,lancedb,weaviate,pinecone: record exceptions and set ERROR status on failed span (#4102)
• langchain: remove orphaned context_api.attach() in on_chain_end (#4100)
• langchain: use positional args for wrap_function_wrapper for wrapt v1/v2 compat, fix BaseChatOpenAI uninstrument (#4082)
• langchain: handle ToolNode in create_react_agent tool iteration (#4081)
• openai: remove redundant try/finally in ResponseStream close and aclose (#4080)
• anthropic: prevent KeyError/IndexError in streaming tool use delta handler (#4079)
• openai: guard against AsyncAPIResponse without .id in async responses wrapper (#4078)
• anthropic: always set streaming token usage from API data (#3949) (#3976)
• instrumentation: langchain association properties context cleanup (#4055)
• evaluator: add reason to PII detector response (#4026)

v0.60.0 (2026-04-19)

Feat

• llamaindex: Instrumentation adjustment for Otel GenAI semconv support (#3979)

Fix

• groq: align instrumentation with OTel GenAI semconv v1.40.0 (#4010)

v0.59.2 (2026-04-16)

Fix

• traceloop-sdk: avoid calling async json methods in JSONEncoder (#3968)

v0.59.1 (2026-04-15)

Fix

• langchain: detach existing SpanHolder token before overwrite in _create_llm_span (#3958)

v0.59.0 (2026-04-13)

Feat

• traceloop-sdk: Add guardrails (#3649)

v0.58.1 (2026-04-12)

Fix

• evaluators: update route api (#3977)

v0.58.0 (2026-04-09)

Feat

• gemini: migrate google-generativeai to latest OTel GenAI semantic conventions (#3840)

v0.57.0 (2026-03-30)

Feat

• bedrock: Instrumentation adjustment for Otel GenAI semconv support (#3845)

v0.56.1 (2026-03-30)

Fix

• SDK + Agno: set gen_ai.tool.name on @tool() decorator and Agno tool spans (#3904)

v0.56.0 (2026-03-30)

Feat

• crewai: trigger version bump for GenAI semconv compliance (#3903)

v0.55.0 (2026-03-29)

Feat

• open-ai: instrumentation to support OTel GenAI Semantic Conventions 0.5.0 (#3844)
• langchain: new semnconv langchain (#3900)

v0.54.0 (2026-03-29)

Feat

• anthropic: conform instrumentation to OTel GenAI semantic conventions (#3835)

v0.53.4 (2026-03-29)

Fix

• release: google_generativeai path in .cz.toml & trigger version bump in all instrumentations (#3896)
• llamaindex: handle None content in StructuredLLM responses (#3513) (#3665)
• semconv: migrating span attributes to OTel gen_ai convention (#3809)

v0.53.3 (2026-03-19)

Fix

• langchain: release and use semconv 0.4.16 version (#3829)

v0.53.2 (2026-03-18)

Fix

• use GITHUB_TOKEN for release creation to resolve 403 error (#3821)

v0.53.1 (2026-03-17)

Fix

• traceloop-sdk: Add dataset override functionality (#3813)

v0.53.0 (2026-03-04)

Feat

• langchain: add OpenTelemetry GenAI semantic conventions (#3673)

Fix

• semconv: revert deleted semconv attributes (#3750)
• pinecone: instrument pinecone package instead of deprecated pinecone-client (#3733)
• langchain: support non-ascii characters to support i18n (#3734)

v0.52.6 (2026-02-26)

Fix

• dataset: Add versions to dataset metadata (#3732)
• qdrant: support all versions of qdrant package (#3500)

v0.52.5 (2026-02-23)

Fix

• traceloop-sdk: Add evaluator config to the evaluator validator (#3706)
• anthropic: restore accidentally lost cache tokens attributes (#3648)

v0.52.4 (2026-02-19)

Fix

• openai-agents: fix realtime session event handling for prompts, completions, and usage (#3688)
• preserve return values for RealtimeSession context manager methods (#3681)
• openai-agents: add functools.wraps to dont_throw decorator (#3687)

v0.52.3 (2026-02-10)

Fix

• openai-agents: add clear flag to support two instrumentation modes (#3489)

v0.52.2 (2026-02-08)

Fix

• traceloop-sdk: Add conversation decorator (#3659)
• traceloop-sdk: Add endpoint_is_traceloop attribute (#3650)

v0.52.1 (2026-02-02)

Fix

• voyageai: add to commitizen to bump on release (#3660)

v0.52.0 (2026-02-02)

Feat

• voyage-ai: add voyage-ai instrumentation (#3653)

Fix

• openai-agents: apply content tracing flag to content (#3487)
• traceloop-sdk: Align evals output schema (#3643)

v0.51.1 (2026-01-26)

Fix

• openai-agents: add support for realtime (#3533)

v0.51.0 (2026-01-20)

Feat

• google-generativeai: Add metrics support (#3506)

Fix

• traceloop-sdk: Add csv and json support to experiment (#3537)
• evals: evals API supports input + config, generate mbt functions (#3534)
• langchain: correct unknown role in completion spans (#3532)
• evals: auto generate evals (#3529)
• tracing: Add association property (#3524)
• openai-agents: optional import of optional deps (#3488)

v0.50.1 (2025-12-16)

Fix

• sample-app: lint fix (#3522)

v0.50.0 (2025-12-15)

Feat

• guardrail: Add guardrail decorator (#3521)

v0.49.8 (2025-12-11)

Fix

• openai: add support for realtime api (websockets) (#3511)
• ollama: support Older Version Ollama (#3501)

v0.49.7 (2025-12-08)

Fix

• exp: Add a real agent example (#3507)
• evals: Add agent evaluators to made by traceloop (#3505)
• exp: Add made by traceloop evaluators (#3503)
• traceloop-sdk: Fixes gRPC exporter initialisation with insecure OTLP (#3481)

v0.49.6 (2025-12-01)

Fix

• agno: add streaming support for Agent.run() and Agent.arun() (#3483)

v0.49.5 (2025-11-27)

Fix

• openai: responses instrumentation broken traces for async streaming (#3475)
• mcp: remove faulty logic of trying to deduce HTTP errors (#3477)

v0.49.4 (2025-11-27)

Fix

• exp: Add run in github experiment (#3459)

v0.49.3 (2025-11-26)

Fix

• openai: recognize NOT_GIVEN and Omit (#3473)
• dataset: add support for file cells in datasets with upload and external URL linking capabilities (#3462)
• openai: report request attributes in responses API instrumentation (#3471)
• sdk: crewai tracing provider conflict (#3470)
• sdk: watsonx warning on initialization (#3469)
• traceloop-sdk: add type-checking support with mypy (#3463)

v0.49.2 (2025-11-25)

Fix

• sdk: remove posthog (#3466)

v0.49.1 (2025-11-24)

Fix

• langchain: allow configuration of metadata key prefix (#3367)
• openai: record service_tier attribute (#3458)

v0.49.0 (2025-11-23)

Feat

• agno: add instrumentation for agno framework (#3452)

v0.48.2 (2025-11-23)

Fix

• add structured outputs schema logging for Anthropic and Gemini (#3454)
• openai: use SpanAttributes instead of GenAIAttributes for cache token attributes (#3442)
• migrate from events api to log records for otel 1.37.0+ compatibility (#3453)

v0.48.1 (2025-11-17)

Fix

• openai: safe handle None tools value in responses api (#3447)
• mcp: move exporter dependency to dev and test environment (#3445)

v0.48.0 (2025-11-11)

Feat

• instrumentation: updated GenAI attributes to use OTel's (#3138)

Fix

• openai: add streaming support for responses.create() api (#3437)
• bedrock: handle non-text contentBlockDelta events in converse_stream (#3404)
• openai-agents: span attribute handling for tool calls and results (#3422)
• watson: collect prompt content and set as span attribute (#3417)

v0.47.5 (2025-10-24)

Fix

• google-genai: make streaming responses work (again) (#3421)
• langchain: changed dictionary access from spans[run_id] to spans.get(run_id) (#3403)

v0.47.4 (2025-10-22)

Fix

• fastmcp: Remote MCP instrumentation (#3419)

v0.47.3 (2025-09-21)

Fix

• openai-agents: propagate gen_ai.agent.name through an agent flow + set workflow name to fast mcp (#3388)

v0.47.2 (2025-09-17)

Fix

• mcp: add mcp.server parent span wrapper for FastMCP tool calls (#3382)

v0.47.1 (2025-09-14)

Fix

• mcp: better instrumentation for FastMCP (#3372)
• anthropic: preserve streaming helper methods in instrumentation (#3377)
• cohere: add v2 api instrumentation (#3378)
• mistralai: instrumentation for version 1.9+ compatibility (#3376)
• sdk: dual bearer send via httpx (#3373)

v0.47.0 (2025-09-10)

Feat

• writer: initial implementation (#3209)

Fix

• crewai: Update CrewAI instrumentation name (#3363)
• sample-app: Update google genai package (#3358)
• traceloop-sdk: include telemetry SDK attributes in tracing (#3359)
• sdk: get default span processor don't work without a base URL (#3360)

v0.46.2 (2025-08-29)

Fix

• vertexai: add missing role attributes when handling images (#3347)
• sdk: manual logging example + fix span ended error (#3352)
• sdk: support disabling all instrumentations (#3353)
• openai-agents: support json inputs (#3354)
• openai: reasoning jsons weren't stored
• crewai: fix unpack error when metrics are disabled (#3345)
• milvus: Set default values when metrics are disabled (#3344)

v0.46.1 (2025-08-24)

Fix

• google-generativeai,vertexai: image support for Gemini models (#3340)

v0.46.0 (2025-08-24)

Feat

• openai: add reasoning attributes (#3336)
• semantic-conventions-ai: Add reasoning attributes (#3330)
• experiment: Add run experiment capabilities (#3331)

Fix

• traceloop-sdk: bump logging instrumentation to support newer otel versions (#3339)
• traceloop-sdk: add @staticmethod decorator to set_association_properties (#3341)
• google-genai: update logic for deciding whether to use awrap or wrap in the Google Generative AI Instrumentation (#3329)
• ollama: missing response model attr in operation duration metric (#3328)
• bedrock: add guardrail on span attributes (#3326)

v0.45.6 (2025-08-18)

Fix

• anthropic: fix with_raw_response wrapper consistency and re-enable beta API instrumentation (#3297)
• langchain: include content attribute when assistant messages have tool calls (#3287)
• google-genai: migrate Google Generative AI instrumentation to googleapis/python-genai (#3282)

v0.45.5 (2025-08-15)

Fix

• openai-agents: switch to hook-based instrumentation (#3283)

v0.45.4 (2025-08-14)

Fix

• relax opentelemetry-semantic-conventions-ai deps (#3259)

v0.45.3 (2025-08-14)

Fix

• anthropic: temp disable beta apis instrumentation (#3258)

v0.45.2 (2025-08-14)

Fix

• langchain: langgraph application crash due to context detach (#3256)

v0.45.1 (2025-08-13)

Fix

• langchain: context detach exception (#3255)
• mcp: MCP Instrumentation: streamablehttp_client Parameter Corruption (#3199)

v0.45.0 (2025-08-12)

Feat

• datasets: add dataset and datasets functionality (#3247)

Fix

• anthropic: support with_raw_response wrapper for span generation (#3250)
• langchain: fix nesting of langgraph spans (#3206)
• langchain: Add "dont_throw" to "on_llm_end" and remove blank file (#3232)

v0.44.3 (2025-08-12)

Fix

• sdk: avoid initializing metrics exporter on custom tracing config (#3249)
• openai: propagate span IDs properly to events (#3243)

v0.44.2 (2025-08-11)

Fix

• openai: dynamically import types for 1.99 (#3244)
• langchain: Added new method for fetching model name from association metadata (#3237)

v0.44.1 (2025-08-04)

Fix

• mcp: do not override meta pydantic types (#3179)

v0.44.0 (2025-08-03)

Feat

• sdk: support multiple span processors (#3207)
• semantic-conentions-ai: add LLMVendor enum to semantic conventions (#3170)

Fix

• langchain: spans dictionary memory leak (#3216)
• openai-agents: use framework's context to infer trace (#3215)
• sdk: respect truncation otel environment variable (#3212)
• anthropic: async stream manager (#3220)
• langchain: populate metadata as span attributes in batch operations (#3218)
• anthropic: various fixes around tools parsing (#3204)
• qdrant: fix qdrant-client auto instrumentation condition (#3208)
• instrumentation: remove param enrich_token_usage and simplify token calculation (#3205)
• langchain: ensure llm spans are created for sync cases (#3201)
• openai: support for openai non-consumed streams (#3155)

v0.43.1 (2025-07-23)

Fix

• langchain: added vendors to llm calls (#3165)

v0.43.0 (2025-07-22)

Feat

• prompts: add tool function support (#3153)

Fix

• llamaindex: structured llm model and temperature parsing (#3159)
• langchain: report token usage histogram (#3059)
• openai: prioritize api-provided token over tiktoken calculation (#3142)
• milvus: Add metrics support (#3013)

v0.42.0 (2025-07-17)

Feat

• llamaindex: support llamaparse instrumentation (#3103)
• milvus: add semantic convention for Milvus DB metrics (#3015)

Fix

• openai-agents: fix broken traces with agents handoff on run_stream (#3143)
• traceloop-sdk: redefine histogram bucket boundaries (#3129)

v0.41.0 (2025-07-13)

Feat

• openai-agents: initial instrumentation; collect OpenAI agent traces and metrics (#2966)
• google-generativeai: implement emitting events in addition to current behavior (#2887)
• vertexai: implement emitting events in addition to current behavior (#2942)
• langchain: implement emitting events in addition to current behavior (#2889)
• anthropic: implement emitting events in addition to current behavior (#2884)
• bedrock: implement emitting events in addition to current behavior (#2885)
• llamaindex: implement emitting events in addition to current behavior (#2941)
• watsonx: implement emitting events in addition to current behavior (#2896)
• cohere: implement emitting events in addition to current behavior (#2886)
• groq: implement emitting events in addition to current behavior (#2888)
• sagemaker: implement emitting events in addition to current behavior (#2894)
• together: implement emitting events in addition to current behavior (#2895)
• replicate: implement emitting events in addition to current behavior (#2893)
• ollama: implement emitting events in addition to current behavior (#2891)
• mistralai: implement emitting events in addition to current behavior (#2890)
• vendor matching (#3062)
• semconv: add an attribute for output schema (#3064)
• transformers: implement the support to emitting events in addition to current behavior (#2940)
• alephalpha: implement emitting events in addition to current behavior (#2880)
• mcp: Add support for mcp streamable http transport type (#3049)
• milvus: Add error.type attribute from OpenTelemetry Semantic Conventions (#3009)
• openai: OpenAI responses minimal instrumentation (#3052)
• ollama: add meter STTG to ollama instrumentation (#3053)
• openai: implement emitting events in addition to current behavior (#2892)

Fix

• align semconv deps (#3106)
• sagemaker: add should_send_prompts checks (#3072)
• watsonx: add should_send_prompts check to model response (#3071)
• groq: add should_send_prompts checks (#3074)
• ollama: add should_send_prompts check (#3073)
• transformers: add should_send_prompts checks (#3070)
• groq: wrong system attribute was given (#3069)
• openai: record exception as span events as well (#3067)
• openai: add request schema attribute (#3065)
• mcp: add support for error_type in mcp instrumentation (#3050)
• google gemini insturmentation (#3055)
• openai: completions.parse out of beta, azure remove double-slash (#3051)

v0.40.14 (2025-06-24)

Fix

• instrumentation dependencies issue for google, ollama and redis (#3044)

v0.40.13 (2025-06-24)

Fix

• sdk: manual report of usage data (#3045)
• sagemaker: Improve _handle_call to safely parse JSON, CSV, and byte inputs (#2963)
• anthropic: serialize assistant message pydantic models (#3041)
• sdk: Ensure instrumentors don’t report successful init if package isn’t installed (#3043)

v0.40.12 (2025-06-20)

Fix

• langchain: add tool call ids to tool message in history (#3033)

v0.40.11 (2025-06-17)

Fix

• sdk: sampling support (#3027)

v0.40.10 (2025-06-17)

Fix

• google-genai: Add support for generate_content method in google genai models (#3014)

v0.40.9 (2025-06-10)

Fix

• langchain: Fix missing langchain dependency for LangGraph tracing (#2988)

v0.40.8 (2025-06-09)

Fix

• openai: dump pydantic input message (#2979)
• langchain: trace langchain tool definitions (#2978)

v0.40.7 (2025-05-20)

Fix

• mcp: Added support for newer version of MCP (#2956)
• gemini: proper chat support (#2948)
• milvus: Add instrumentation for pymilvus MilvusClient hybrid search operation (#2945)

v0.40.6 (2025-05-16)

Fix

• sdk: support overriding the span processor on_end hook (#2947)
• milvus: Added New Semantic Conventions for pymilvus MilvusClient Hybrid Search (#2944)

v0.40.5 (2025-05-13)

Fix

• langchain: tools in message history (#2939)
• sdk: Place MCP in its lexical order (#2943)

v0.40.4 (2025-05-10)

Fix

• milvus: Enhanced Milvus VectorDB Instrumentation for Improved search Monitoring (#2815)
• milvus: Added New Semantic Conventions for Milvus Search (Request for Version Update 0.4.5 -> 0.4.6) (#2883)
• MCP: Added error status to traces in MCP server for tool calls (#2914)
• ollama: pre-imported funcs instrumentation failure (#2871)

v0.40.3 (2025-05-07)

Fix

• langchain: report token counts when trace content is enabled (#2899)
• mcp+anthropic: vanilla mcp crashed due to argument manipulation (#2881)

v0.40.2 (2025-04-30)

Fix

• ci: align mcp instrumentation version (#2876)

v0.40.1 (2025-04-30)

Fix

• ci-cd: add mcp to commitizen (#2875)

v0.40.0 (2025-04-30)

Feat

• instrumentation: Adding MCP opentelemetry-instrumentation into traceloop (#2829)

v0.39.4 (2025-04-28)

Fix

• sdk: improve type safety in decorators (#2867)

v0.39.3 (2025-04-24)

Fix

• langchain: support cached tokens attributes logging (#2830)

v0.39.2 (2025-04-18)

Fix

• openai: add cache read tokens from returned usage block (#2820)
• ollama: type error in dict combination of ollama instrumentation (#2814)
• llama-index: use the correct instrumentation point (#2807)

v0.39.1 (2025-04-15)

Fix

• sdk: Loosen tenacity dependency constraint to allow versions up to 10.0 (#2816)

v0.39.0 (2025-03-25)

Feat

• instrumentation: add metric for Bedrock prompt caching (#2788)
• bedrock: add support for ARN and cross region endpoint (#2785)
• instrumentation: Support Converse APIs and guardrail metrics (#2725)

Fix

• bedrock: add span attr for Bedrock prompt caching (#2789)
• anthropic: add thinking as a separate completion message (#2780)
• langchain: support for date/time in langchain serializations (#2792)
• openai: set user messages as prompts, not completions (#2781)
• groq: exception when metrics are turned off (#2778)
• ollama: Implemented meter in the instrumentation (#2741)

v0.38.12 (2025-03-07)

Fix

• sdk: client shouldn't be initialized if destination is not traceloop (#2754)

v0.38.11 (2025-03-06)

Fix

• sdk: When tracing task with no name provided , use qualified name instaed of name (#2743)

v0.38.10 (2025-03-05)

Fix

• sdk: record exceptions (#2733)

v0.38.9 (2025-03-04)

Fix

• milvus: updated the instrumentation to collect get() and create_collection() span attributes (#2687)
• semconv: added new semantic conventions for milvus db (#2727)

v0.38.8 (2025-02-27)

Fix

• vertexai: support generative model chat session tracing (#2689)
• groq: Updated the instrumentation to collect the token histogram (#2685)

v0.38.7 (2025-02-19)

Fix

• langchain: warning with mixed metadata value types (#2665)
• langchain: handle errors (#2664)
• groq: streaming support (#2663)

v0.38.6 (2025-02-17)

Fix

• sdk: async generator wrapping (#2635)
• instrumentation: watsonx initialize parameters (#2633)

v0.38.5 (2025-02-10)

Fix

• sdk: Fix async decorator input & output json encoder (#2629)

v0.38.4 (2025-02-06)

Fix

• sdk: improve package name detection with type hints and null safety (#2618)

v0.38.3 (2025-02-05)

Fix

• langchain: sanitize metadata (#2608)
• sdk: improve package name detection for metadata (#2607)
• openai: support context propagation for OpenAI v0 and v1 (#2606)
• openai: report exceptions on spans (#2604)

v0.38.2 (2025-02-04)

Fix

• dependency issue for python 3.13 (#2603)
• openai: compatibility issue with openai attribute not available (#2602)

v0.38.1 (2025-02-04)

Fix

• openai: don't eagerly import types from openai as they may not be available (#2601)
• sdk: watsonx package name was wrong preventing instrumentation (#2600)

v0.38.0 (2025-02-04)

Feat

• crewai: Implemented histogram for crewai instrumentation (#2576)

Fix

• mistral: add response id attribute (#2549)
• openai: add response id attribute (#2550)
• together: add response id attribute (#2551)
• langchain: add response id (anthropic only) (#2548)

v0.37.1 (2025-01-25)

Fix

• crewai: instrumentation package version (#2570)

v0.37.0 (2025-01-25)

Feat

• crewai: initial instrumentation; collect agent and task traces (#2489)

Fix

• add crewai to supported frameworks in readme (#2559)
• cohere: add response id attribute (#2545)
• bedrock: add response id attribute (#2544)
• anthropic: add response id attribute (#2543)
• groq: add response id attribute (#2546)
• langchain: address warning for NoneType attibute; update some test deps (#2539)

v0.36.1 (2025-01-20)

Fix

• sdk: remove use of __all__ (#2536)
• gemini: set prompt attributes for role and content (#2511)

v0.36.0 (2025-01-13)

Feat

• sdk: client, annotations (#2452)

v0.35.0 (2025-01-03)

Feat

• sdk: combine async/sync decorators & annotations (#2442)

Fix

• sdk: option to disable SDK (#2454)

v0.34.1 (2024-12-22)

Fix

• llamaindex: workflow context detach exception and span names (#2421)
• langchain: azure openai missing request model (#2416)

v0.34.0 (2024-12-12)

Feat

• docs: Update README.md to add link to recently-added GCP integration document. (#2384)

Fix

• bump otel >0.50b0 (#2386)
• update google generative AI import (#2382)
• sdk: Update JSONEncoder to allow class instance methods to be serializable (#2383)
• openai: Add token count and system to assistants (#2323)
• openai: Add trace context to client requests (#2321)

v0.33.12 (2024-11-13)

Fix

• sdk: aworkflow decorator for async generators (#2292)
• cohere: rerank exception on saving response when return_documents=True (#2289)
• sdk: gemini instrumentation was never installed due to package name error (#2288)

v0.33.11 (2024-11-09)

Fix

• sdk: remove print (#2285)

v0.33.10 (2024-11-08)

Fix

• general bump of otel dependencies (#2274)

v0.33.9 (2024-11-05)

Fix

• openai: exception throw with pydantic v1 (#2262)

v0.33.8 (2024-11-05)

Fix

• vertex: async / streaming was missing output fields (#2253)

v0.33.7 (2024-11-04)

Fix

• sdk: don't serialize large jsons as inputs/outputs (#2252)

v0.33.6 (2024-11-04)

Fix

• anthropic: instrument anthropics system message as gen_ai.prompt.0 (#2238)
• llamaindex: streaming LLMs caused detached spans (#2237)
• sdk: missing sagemaker initialization (#2235)
• sdk: support a "block-list" of things to not instrument (#1958)

v0.33.5 (2024-10-29)

Fix

• openai+anthropic: async call crashing the app when already in a running asyncio loop (#2226)

v0.33.4 (2024-10-28)

Fix

• langchain: structured output response parsing (#2214)
• anthropic: add instrumentation for Anthropic prompt caching (#2175)
• bedrock: cohere models failed to report prompts (#2204)

v0.33.3 (2024-10-22)

Fix

• sdk: capture posthog events as anonymous; roll ingestion key (#2194)

v0.33.2 (2024-10-17)

Fix

• langchain: various bugs and edge cases in metric exporting (#2167)
• sdk: add header for logging exporter (#2164)

v0.33.1 (2024-10-16)

Fix

• langchain: metrics support (#2154)
• langchain: Add trace context to client requests (#2152)
• anthropic: add instrumentation for Anthropic tool calling (alternative to #1372) (#2150)
• openai: add structured output instrumentation (#2111)

v0.33.0 (2024-10-15)

Feat

• sdk: add OpenTelemetry logging support (#2112)
• ollama: tool calling (#2059)

Fix

• llama-index: add attribute to span for llm request type in dispatcher wrapper (#2141)

v0.32.2 (2024-10-04)

Fix

• traceloop-sdk: add aiohttp as dependency (#2094)

v0.32.1 (2024-10-03)

Fix

• anthropic: Replace count_tokens with usage for newer models (#2086)

v0.32.0 (2024-10-03)

Feat

• bedrock: support metrics for bedrock (#1957)
• SageMaker: Add SageMaker instrumentation (#2028)

Fix

• langchain: token usage reporting (#2074)

v0.31.4 (2024-10-01)

Fix

• langchain: serialize inputs and outputs with pydantic (#2065)
• sdk: custom image uploader (#2064)

v0.31.3 (2024-09-28)

Fix

• support async image upload flows (#2051)

v0.31.2 (2024-09-27)

Fix

• anthropic: add support for base64 images upload for anthropic (#2029)

v0.31.1 (2024-09-26)

Fix

• anthropic: token counting exception when prompt contains images (#2030)

v0.31.0 (2024-09-25)

Feat

• sdk+openai: support base64 images upload (#2000)

Fix

• sdk: wrong package check for Vertex AI (#2015)

v0.30.1 (2024-09-16)

Fix

• langchain: support v0.3.0 (#1985)

v0.30.0 (2024-09-04)

Feat

• add groq instrumentation (#1928)

v0.29.2 (2024-08-31)

Fix

• langchain: allow external and langchain metadata (#1922)

v0.29.1 (2024-08-29)

Fix

• bedrock: llama3 completion wasnt logged (#1914)

v0.29.0 (2024-08-29)

Feat

• instrumentation: Import redis from OpenTelemetry, add redis sample rag application (#1837)

Fix

• langchain: add missing kind property (#1901)

v0.28.2 (2024-08-26)

Fix

• openai: calculating streaming usage didnt work on azure models

v0.28.1 (2024-08-24)

Fix

• langchain: langgraph traces were broken (#1895)

v0.28.0 (2024-08-24)

Feat

• llama-index: callback improvements (#1859)

Fix

• openai: re-enabled token count for azure instances (#1877)
• openai: not given values thrown errors (#1876)
• sdk: aentity_class was missing a positional argument (#1816)
• sdk: instrument threading for propagating otel context (#1868)
• openai: TypeError: '<' not supported between instances of 'NoneType' and 'int' in embeddings_wrappers.py (#1836)

v0.27.0 (2024-08-15)

Feat

• llama-index: Use callbacks (#1546)
• LanceDB Integration (#1749)
• sdk: chained entity path on nested tasks (#1782)

Fix

• workflow_name and entity_path support for langchain + fix entity_name (#1844)
• sdk: disable traceloop sync by default (#1835)

v0.26.5 (2024-08-06)

Fix

• langchain: export metadata as association properties (#1805)
• bedrock: add model name for amazon bedrock response (#1757)

v0.26.4 (2024-08-03)

Fix

• bedrock: token count for titan (#1748)

v0.26.3 (2024-08-02)

Fix

• langchain: various cases where not all parameters were logged properly (#1725)

v0.26.2 (2024-07-31)

Fix

• separate semconv-ai module to avoid conflicts (#1716)

v0.26.1 (2024-07-30)

Fix

• bump to otel 0.47b0 (#1695)
• openai: log content filter results in proper attributes (#1539)

v0.26.0 (2024-07-26)

Feat

• openai: add tool call id (#1664)

Fix

• pinecone: support v5 (#1665)

v0.25.6 (2024-07-23)

Fix

• sdk: aworkflow wasn't propagating workflow_name attribute (#1648)
• langchain: agent executor weren't producing traces (#1616)

v0.25.5 (2024-07-17)

Fix

• openai: pydantic tool calls in prompt weren't serialized correctly (#1572)

v0.25.4 (2024-07-15)

Fix

• sdk: manual reporting of llm spans (#1555)

v0.25.3 (2024-07-11)

Fix

• langchain: input/output values weren't respecting user config (#1540)

v0.25.2 (2024-07-11)

Fix

• llamaindex: report entity name (#1525)
• langchain: remove leftover print
• langchain: cleanups, and fix streaming issue (#1522)
• langchain: report llm spans (instead of normal instrumentations) (#1452)

v0.25.1 (2024-07-09)

Fix

• association properties and workflow / task on metrics (#1494)
• llamaindex: report inputs+outputs on entities (#1495)

v0.25.0 (2024-07-08)

Feat

• suppress LLM instrumentations through context (#1453)
• langchain: improve callbacks (#1426)

Fix

• sdk: llamaindex instrumentation was never initialized (#1490)

v0.24.0 (2024-07-03)

Feat

• sdk: prompt versions and workflow versions (#1425)
• openai: add support for parallel function calls (#1424)
• marqo: Add marqo instrumentation (#1373)

Fix

• sdk: context detach issues on fastapi (#1432)
• openai: Handle tool_calls assistant messages (#1429)
• sdk: speedup SDK initialization (#1374)
• gemini: relax version requirements (#1367)

Refactor

• openai: rename function_call to tool_calls (#1431)

v0.23.0 (2024-06-17)

Feat

• langchain: use callbacks (#1170)

Fix

• input/output serialization issue for langchain (#1341)
• sdk: remove auto-create dashboard option (#1315)

v0.22.1 (2024-06-13)

Fix

• sdk: backpropagate association property to nearest workflow/task (#1300)
• sdk: clear context when @workflow or @task is ending (#1301)
• bedrock: utilize invocation metrics from response body for AI21, Anthropic, Meta models when available to record usage on spans (#1286)

v0.22.0 (2024-06-10)

Feat

• gemini: basic support in generate_content API (#1293)
• alephalpha: Add AlephAlpha instrumentation (#1285)
• instrumentation: add streamed OpenAI function tracing (#1284)
• togetherai: Add together ai instrumentation (#1264)

Fix

• anthropic: duplicate creation of metrics (#1294)
• haystack: add input and output (#1202)
• openai: calculate token usage for azure (#1274)
• use constants (#1131)
• instrumentation: Handle OpenAI run polling (#1256)

v0.21.5 (2024-06-05)

Fix

• openai: handle empty finish_reason (#1236)
• removed debug prints from instrumentations
• vertexai: change the span names to match method calls (#1234)

v0.21.4 (2024-06-03)

Fix

• openai+anthropic+watsonx: align duration and token.usage metrics attributes with conventions (#1182)

v0.21.3 (2024-06-03)

Fix

• openai: async streaming responses (#1229)
• sdk: temporarily (?) remove sentry (#1228)

v0.21.2 (2024-05-31)

Fix

• all packages: Bump opentelemetry-api to 1.25.0 and opentelemetry-instrumentation to 0.46b0 (#1189)

v0.21.1 (2024-05-30)

Fix

• log tracing errors on debug level (#1180)
• bedrock: support streaming API (#1179)
• weaviate: support v4.6.3 (#1134)
• sdk: wrong package check for mistral instrumentations (#1168)

v0.21.0 (2024-05-27)

Feat

• vertexai: vertexai.generative_models / llm_model detection (#1141)

Fix

• bedrock: support simple string in prompts (#1167)
• langchain: stringification fails for lists of LangChain Documents (#1140)

v0.20.0 (2024-05-26)

Feat

• mistral: implement instrumentation (#1139)
• ollama: implement instrumentation (#1138)

Fix

• anthropic: don't fail if can't count anthropic tokens (#1142)
• ollama: proper unwrapping; limit instrumentations to versions <1
• bedrock: instrument bedrock calls for Langchain (with session) (#1135)

v0.19.0 (2024-05-22)

Feat

• milvus: add Milvus instrumentation (#1068)

Fix

• add explicit buckets to pinecone histograms (#1129)
• pinecone: backport to v2.2.2 (#1122)
• llm metrics naming + views (#1121)
• langchain: better serialization of inputs and outputs (#1120)
• sdk: failsafe against instrumentation initialization errors (#1117)
• sdk: instrument milvus (#1116)

v0.18.2 (2024-05-17)

Fix

• openai: old streaming handling for backward compatibility with OpenAI v0 (#1064)
• openai: report fingerprint from response (#1066)
• sdk: special handling for metrics with custom traces exporter (#1065)

v0.18.1 (2024-05-17)

Fix

• openai: fallback to response model if request model is not set when calculating token usage (#1054)
• openai: add default value of stream as false in token usage metric (#1055)

v0.18.0 (2024-05-14)

Feat

• pinecone: metrics support (#1041)

Fix

• sdk: handle workflow & tasks generators (#1045)
• cohere: use billed units for token usage (#1040)

v0.17.7 (2024-05-13)

Fix

• remove all un-needed tiktoken deps (#1039)

v0.17.6 (2024-05-13)

Fix

• sdk: removed unneeded tiktoken dependency (#1038)

v0.17.5 (2024-05-13)

Fix

• openai: relax tiktoken requirements (#1035)

v0.17.4 (2024-05-13)

Fix

• sdk: loosen SDK requirements for Sentry + Posthog (#1027)

v0.17.3 (2024-05-08)

Fix

• sdk: separate sentry SDK (#1004)

v0.17.2 (2024-05-07)

Fix

• langchain: support model-specific packages (#985)
• pinecone: filter argument may be dict (#984)

v0.17.1 (2024-05-01)

Fix

• instrumentation: correct the module declaration to match package filepath name (#940)

v0.17.0 (2024-04-29)

Feat

• sdk: otel metrics with traceloop (#883)
• Updated semantic conventions based on otel community (#884)

Fix

• sdk: do not instrument sentry requests (used internally by SDK) (#939)

v0.16.9 (2024-04-26)

Fix

• openai: missing await for Embedding.acreate (#900)
• cohere: support v5 (#899)
• pinecone: support v3 (#895)
• instrumentation: the build problem for watsonx auto instrumentation (#885)

v0.16.8 (2024-04-25)

Fix

• langchain: input/output reporting (#894)
• sdk: reset the color of messages in the custom metrics exporter (#893)

v0.16.7 (2024-04-25)

Fix

• openai: azure filtering masked all completions (#886)
• chromadb: exception thrown when metadata isn't set (#882)

v0.16.6 (2024-04-19)

Fix

• properly handle and report exceptions (#748)
• langchain: bug when retrieving messages as kwargs from model invoke (#856)
• openai: handle filtered content (#854)
• bedrock: loosen version requirement of anthropic (#830)
• haystack: V2 Support (#710)

v0.16.5 (2024-04-17)

Fix

• sdk: warn for reporting score when not using Traceloop (#829)
• openai: fix aembeddings init error (#828)
• openai: missing aembedding metrics

v0.16.4 (2024-04-15)

Fix

• anthropic: fix issue with disabled metrics (#820)

v0.16.3 (2024-04-15)

Fix

• openai: missing metrics for OpenAI v0 instrumentation (#818)

v0.16.2 (2024-04-14)

Fix

• bedrock: enrich token usage for anthropic calls (#805)
• langchain: use chain names if exist (#804)

v0.16.1 (2024-04-11)

Fix

• llamaindex: proper support for custom LLMs (#776)
• anthropic: prompt attribute name (#775)
• langchain: BedrockChat model name should be model_id (#763)

v0.16.0 (2024-04-10)

Feat

• instrumentation-anthropic: Support for OpenTelemetry metrics for Anthropic (#764)

Fix

• bedrock: support anthropic v3 (#770)

v0.15.13 (2024-04-08)

Fix

• sdk: custom instruments missing parameters (#769)
• sdk: import of removed method
• sdk: removed deprecated set_context

v0.15.12 (2024-04-08)

Fix

• anthropic: do not fail for missing methods
• anthropic: Async and streaming Anthropic (#750)

v0.15.11 (2024-04-04)

Fix

• openai: async streaming metrics (#749)

v0.15.10 (2024-04-04)

Fix

• anthropic: token usage (#747)

v0.15.9 (2024-04-03)

Fix

• openai: switch to init flag for token usage enrichment (#745)
• anthropic: support multi-modal (#746)
• langchain: instrument chat models (#741)

v0.15.8 (2024-04-03)

Fix

• bump otel -> 0.45.0 (#740)

v0.15.7 (2024-04-03)

Fix

• enrich spans with related entity name + support entities nesting (#713)

v0.15.6 (2024-04-02)

Fix

• sdk: stricter dependencies for instrumentations

v0.15.5 (2024-04-02)

Fix

• openai: missing metric for v0 instrumentation (#735)

v0.15.4 (2024-03-31)

Fix

• traceloop-sdk: default value for metrics endpoint (#711)

v0.15.3 (2024-03-28)

Fix

• instrumentation deps without the SDK (#707)
• langchain: support custom models (#706)

v0.15.2 (2024-03-27)

Fix

• openai: enrich assistant data if not available (#705)

v0.15.1 (2024-03-27)

Fix

• openai: support pre-created assistants (#701)

v0.15.0 (2024-03-26)

Feat

• openai: assistants API (#673)
• pinecone: instrument pinecone query embeddings (#368)

Fix

• traceloop-sdk: custom span processor's on_start is honored (#695)
• openai: do not import tiktoken if not used
• sdk: exclude api.traceloop.com from requests
• openai: Support report token usage in stream mode (#661)

v0.14.5 (2024-03-21)

Fix

• anthropic: support messages API (#671)

v0.14.4 (2024-03-21)

Fix

• auto-instrumentation support (#662)
• sample: poetry issues; litellm sample
• sdk: better logging for otel metrics
• sdk: error for manually providing instrumentation list

v0.14.3 (2024-03-17)

Fix

• support python 3.12 (#639)
• traceloop-sdk: Log error message when providing wrong API key. (#638)

v0.14.2 (2024-03-15)

Fix

• openai: support tool syntax (#630)

v0.14.1 (2024-03-12)

Fix

• sdk: protect against unserializable inputs/outputs (#626)

0.14.0 (2024-03-12)

Feat

• watsonx instrumentation: Watsonx metric support (#593)

Fix

• instrumentations: add entry points to support auto-instrumentation (#592)

0.13.3 (2024-03-07)

Fix

• llamaindex: backport to support v0.9.x (#590)
• openai: is_streaming attribute (#589)

0.13.2 (2024-03-06)

Fix

• openai: span events on completion chunks in streaming (#586)
• openai: streaming metrics (#585)

0.13.1 (2024-03-01)

Fix

• watsonx: Watsonx stream generate support (#552)
• watsonx instrumentation: Init OTEL_EXPORTER_OTLP_INSECURE before import watsonx models (#549)
• link back to repo in pyproject.toml (#548)

0.13.0 (2024-02-28)

Feat

• basic Support for OpenTelemetry Metrics and Token Usage Metrics in OpenAI V1 (#369)
• weaviate: implement weaviate instrumentation (#394)

Fix

• watsonx: exclude http request, adding span for model initialization (#543)

0.12.5 (2024-02-27)

Fix

• llamaindex: instrument agents & tools (#533)
• openai: Fix with_raw_response redirect crashing span (#536)
• openai: track client attributes for v1 SDK of OpenAI (#522)
• sdk: replaced MySQL instrumentor with SQLAlchemy (#531)

0.12.4 (2024-02-26)

Fix

• sdk: fail gracefully if input/output is not json serializable (#525)

0.12.3 (2024-02-26)

Fix

• new PR template (#524)

0.12.2 (2024-02-23)

Fix

• cohere: enrich rerank attributes (#476)

0.12.1 (2024-02-23)

Fix

• llamaindex: support query pipeline (#475)

0.12.0 (2024-02-22)

Feat

• Qdrant instrumentation (#364)

Fix

• langchain: support LCEL (#473)
• sdk: fail gracefully in case input/output serialization failure (#472)

0.11.3 (2024-02-19)

Fix

• llamaindex: support both new and legacy llama_index versions (#422)

0.11.2 (2024-02-16)

Fix

• sdk: url for getting API key (#424)

0.11.1 (2024-02-14)

Fix

• openai: handle async streaming responses for openai v1 client (#421)

0.11.0 (2024-02-13)

Feat

• support both new and legacy llama_index versions (#420)

Fix

• sdk: support input/output of tasks & workflows (#419)

0.10.5 (2024-02-13)

Fix

• langchain: backport to 0.0.346 (#418)

0.10.4 (2024-02-08)

Fix

• openai: handle OpenAI async completion streaming responses (#409)

0.10.3 (2024-01-30)

Fix

• README

0.10.2 (2024-01-25)

Fix

• re-enabled haystack instrumentation (#77)

0.10.1 (2024-01-24)

Fix

• resource_attributes always being None (#359)

0.10.0 (2024-01-22)

Feat

• watsonx support for traceloop (#341)

Fix

• sdk: support arbitrary resources (#338)

0.9.4 (2024-01-15)

Fix

• bug in managed prompts (#337)

0.9.3 (2024-01-15)

Fix

• support langchain v0.1 (#320)

0.9.2 (2024-01-12)

Fix

• otel deps (#336)

0.9.1 (2024-01-12)

Fix

• openai: instrument embeddings APIs (#335)

0.9.0 (2024-01-11)

Feat

• google-vertexai-instrumentation (#289)

0.8.2 (2024-01-10)

Fix

• version bump error with replicate (#318)
• version bump error with replicate (#318)

0.8.1 (2024-01-10)

Fix

• replicate release (#316)

0.8.0 (2024-01-04)

Feat

• semconv: added top-k (#291)

Fix

• support anthropic v0.8.1 (#301)
• ci: fix replicate release (#285)

0.7.0 (2023-12-21)

Feat

• replicate support (#248)

Fix

• support pydantic v1 (#282)
• broken tests (#281)

0.6.0 (2023-12-16)

Feat

• sdk: user feedback scores (#247)

0.5.3 (2023-12-12)

Fix

• openai: async streaming instrumentation (#245)

0.5.2 (2023-12-09)

Fix

• send SDK version on fetch requests (#239)

0.5.1 (2023-12-08)

Fix

• support async workflows in llama-index and openai (#233)

0.5.0 (2023-12-07)

Feat

• sdk: support vision api for prompt management (#234)

0.4.2 (2023-12-01)

Fix

• openai: langchain streaming bug (#225)

0.4.1 (2023-11-30)

Fix

• traceloop-sdk: support explicit prompt versioning in prompt management (#221)

0.4.0 (2023-11-29)

Feat

• bedrock support (#218)

Fix

• lint issues

0.3.6 (2023-11-27)

Fix

• openai: attributes for functions in request (#211)

0.3.5 (2023-11-23)

Fix

• llama-index: support ollama completion (#212)

0.3.4 (2023-11-22)

Fix

• sdk: flag for dashboard auto-creation (#210)

0.3.3 (2023-11-22)

Fix

• new logo

0.3.2 (2023-11-16)

Fix

• python 3.8 compatibility (#198)
• cohere: cohere chat token usage (#196)

0.3.1 (2023-11-14)

Fix

• disable telemetry in tests (#171)

0.3.0 (2023-11-10)

Feat

• sdk telemetry data (#168)

Fix

• make auto-create path persisted (#170)

0.2.1 (2023-11-09)

Fix

• openai: yield chunks for streaming (#166)

0.2.0 (2023-11-08)

Feat

• llamaindex auto instrumentation (#157)

0.1.12 (2023-11-07)

Fix

• openai: new OpenAI API v1 (#154)

0.1.11 (2023-11-06)

Fix

• sdk: max_tokens are now optional from the backend (#153)

0.1.10 (2023-11-03)

Fix

• errors on logging openai streaming completion calls (#144)

0.1.9 (2023-11-03)

Fix

• langchain: improved support for agents and tools with Langchain (#143)
• support streaming API for OpenAI (#142)

0.1.8 (2023-11-02)

Fix

• prompt-registry: remove redundant variables print

0.1.7 (2023-11-01)

Fix

• tracing: add missing prompt manager template variables to span attributes (#140)

0.1.6 (2023-11-01)

Fix

• sdk: allow overriding processor & propagator (#139)
• proper propagation of api key to fetcher (#138)

0.1.5 (2023-10-31)

Fix

• ci-cd: release workflow fetches the outdated commit on release package jobs

0.1.4 (2023-10-31)

Fix

• disable syncing when no API key is defined (#135)
• ci-cd: finalize release flow (#133)

0.1.3 (2023-10-30)

Fix

• ci-cd: fix release workflow publish step

0.1.2 (2023-10-30)

Fix

• ci-cd: fix release workflow publish step

0.1.1 (2023-10-30)

Fix

• ci-cd: fix release workflow publish step

0.1.0 (2023-10-30)

Feat

• ci-cd: add release workflow (#132)

Fix

• release workflow credentials

v0.0.70 (2023-10-29)

Feat

• disable content tracing for privacy reasons (#118)
• add prompt version hash (#119)
• propagate prompt management attributes to llm spans (#109)
• support association IDs as objects (#111)
• hugging-face transformers pipeline instrumentation (#104)
• add chromadb instrumentation + fix langchain instrumentation (#103)
• export to Grafana tempo (#95)
• langchain instrumentation (#88)
• cohere: support for chat and rerank (#84)
• cohere instrumentation (#82)
• Anthropic instrumentation (#71)
• basic prompt management (#69)
• Pinecone Instrumentation (#3)
• basic testing framework (#70)
• haystack instrumentations (#55)
• auto-create link to traceloop dashboard
• setting headers for exporting traces
• sdk code + openai instrumentation (#4)

Fix

• sdk: disable sync when using external exporter
• disable content tracing when not overridden (#121)
• langchain: add retrieval_qa workflow span (#112)
• traceloop-sdk: logging of service name in traces (#99)
• do not trigger dashboard auto-creation if exporter is set (#96)
• docs: clarification on getting API key
• chore: spaces and nits on README
• docs: bad link for python SDK
• docs: updated TRACELOOP_BASE_URL (#81)
• add openai function call data to telemetry (#80)
• sdk: disabled prompt registry by default (#78)
• support pinecone non-grpc (#76)
• support python 3.12
• docs: upgrades; docs about prompt mgmt (#74)
• traceloop-sdk: missing lockfile (#72)
• traceloop-sdk: flushing in notebooks (#66)
• py security issue
• docs: update exporting.mdx to include nr instrumentation (#12)
• sdk: async decorators not awaited
• sdk: missing dependency
• warn if Traceloop wasn't initialized properly (#11)
• match new dashboard API
• traceloop-sdk: duplicate spans reporting (#10)
• moved api key to /tmp
• /v1/traces is always appended to endpoint
• parse headers correctly
• traceloop-sdk: replace context variables with otel context + refactor (#8)
• traceloop sdk initialization and initial versions release for instrumentations (#7)
• wrong imports and missing code components (#6)
• gitignore
• README