Supported Integrations
LLM providers, frameworks, and vector stores auto-instrumented by the SDK, by language.
Supported Integrations
The SDK auto-instruments the libraries below — calls are captured as traces with no manual
trace() / generation() code. Support varies by language and reflects what each SDK hooks at
runtime.
To enable any of these, install the library and initialize the SDK first — see Auto-Instrumentation. Provider-specific guides live under Integrations.
Legend: ✅ Supported · — Not available for this language. Newer integrations require a recent SDK version. Regardless of this table, you can record any provider's calls with the manual tracing API.
LLM Providers
| Provider | TypeScript | Python | Java |
|---|---|---|---|
| OpenAI | ✅ | ✅ | ✅ |
| Anthropic | ✅ | ✅ | ✅ |
| Azure OpenAI 1 | ✅ | ✅ | — |
| Google Gemini (AI Studio) 2 | ✅ | ✅ | ✅ |
| Vertex AI | ✅ | ✅ | — |
| Amazon Bedrock | ✅ | ✅ | ✅ |
| LiteLLM | — | ✅ | — |
1 Traced via the OpenAI client configured against an Azure endpoint.
2 Package differs by language: @google/genai (Node), google-generativeai (Python), google-genai (Java).
Frameworks & Agent Frameworks
| Framework | TypeScript | Python | Java |
|---|---|---|---|
| LangChain (incl. LangGraph) 3 | ✅ | ✅ | — |
| LangChain4j | — | — | ✅ |
| Spring AI | — | — | ✅ |
| LlamaIndex | ✅ | ✅ | — |
| Vercel AI SDK 4 | ✅ | — | — |
| Google ADK | ✅ | ✅ | — |
| OpenAI Agents | ✅ | ✅ | — |
| Strands Agents | ✅ | ✅ | — |
| Claude Agent SDK | ✅ | ✅ | — |
| Mastra | ✅ | — | — |
| Amazon Bedrock AgentCore | ✅ | ✅ | — |
| CrewAI | — | ✅ | — |
| AutoGen | — | ✅ | — |
| DSPy | — | ✅ | — |
| Pydantic AI | — | ✅ | — |
| AgentScope | — | ✅ | — |
| ModelScope Agent | — | ✅ | — |
| Microsoft Agent Framework | — | ✅ | — |
3 On Java, use LangChain4j (listed separately).
4 Traced by wrapping the ai module with wrapVercelAI(). See Use with Vercel AI SDK.
Vector Stores
| Store | TypeScript | Python | Java |
|---|---|---|---|
| Pinecone | ✅ | — | ✅ |
| pgvector 5 | ✅ | ✅ | — |
5 On Node, both the pg and postgres (postgres.js) drivers are instrumented.
Other
| Integration | TypeScript | Python | Java |
|---|---|---|---|
| Hugging Face | ✅ | ✅ | — |
| HTTP web frameworks 6 | — | ✅ | — |
6 Python instruments Flask, FastAPI, Django, and Starlette so LLM calls nest under the incoming request. See HTTP Instrumentation.
Provider guides
Use with Anthropic
Record Anthropic prompt-caching tokens (cache_read / cache_creation) and cost in your traces.
Use with Vercel AI SDK
Trace generateText, streamText, embeddings, tools, and agents by wrapping the ai module.
See also
- Auto-Instrumentation — install and enable per provider/framework
- Manual Tracing — works with any provider, instrumented or not