> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bryel.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Integrations

> Every provider SDK, agent SDK, and framework bryel captures — and the one line that turns each on.

bryel is OpenTelemetry-native, so it traces any library that has an
[OpenInference](https://github.com/Arize-ai/openinference) instrumentor. You
don't map anything — install the instrumentor for your stack and bryel exports
the spans.

In **Python**, one call wires up every instrumentor you have installed:

```python theme={"theme":{"light":"github-light","dark":"github-dark"}}
import bryel
bryel.init(service_name="my-app", instrument=True)  # reads BRYEL_KEY
```

`instrument=True` scans for the libraries below and turns on the ones present.
Prefer to be explicit? Call a single instrumentor yourself — the table shows how.

<Note>In **TypeScript/JavaScript**, add any OpenTelemetry GenAI instrumentation to your `NodeSDK` and ship it with `bryelSpanProcessor` — see [TypeScript](/sdk/typescript#provider-sdks). Any other language can send over the [HTTP API](/sdk/http).</Note>

## Provider SDKs

| Provider       | Package                                      | Instrumentor              |
| -------------- | -------------------------------------------- | ------------------------- |
| OpenAI         | `openinference-instrumentation-openai`       | `OpenAIInstrumentor`      |
| Anthropic      | `openinference-instrumentation-anthropic`    | `AnthropicInstrumentor`   |
| Amazon Bedrock | `openinference-instrumentation-bedrock`      | `BedrockInstrumentor`     |
| Google GenAI   | `openinference-instrumentation-google-genai` | `GoogleGenAIInstrumentor` |
| VertexAI       | `openinference-instrumentation-vertexai`     | `VertexAIInstrumentor`    |
| Groq           | `openinference-instrumentation-groq`         | `GroqInstrumentor`        |
| Mistral        | `openinference-instrumentation-mistralai`    | `MistralAIInstrumentor`   |
| LiteLLM        | `openinference-instrumentation-litellm`      | `LiteLLMInstrumentor`     |
| Instructor     | `openinference-instrumentation-instructor`   | `InstructorInstrumentor`  |

## Agent SDKs

| SDK           | Package                                       | Instrumentor               |
| ------------- | --------------------------------------------- | -------------------------- |
| OpenAI Agents | `openinference-instrumentation-openai-agents` | `OpenAIAgentsInstrumentor` |
| Google ADK    | `openinference-instrumentation-google-adk`    | `GoogleADKInstrumentor`    |

## Frameworks

| Framework             | Package                                     | Instrumentor             |
| --------------------- | ------------------------------------------- | ------------------------ |
| LangChain / LangGraph | `openinference-instrumentation-langchain`   | `LangChainInstrumentor`  |
| LlamaIndex            | `openinference-instrumentation-llama-index` | `LlamaIndexInstrumentor` |
| CrewAI                | `openinference-instrumentation-crewai`      | `CrewAIInstrumentor`     |
| DSPy                  | `openinference-instrumentation-dspy`        | `DSPyInstrumentor`       |
| Haystack              | `openinference-instrumentation-haystack`    | `HaystackInstrumentor`   |
| smolagents            | `openinference-instrumentation-smolagents`  | `SmolagentsInstrumentor` |
| Agno                  | `openinference-instrumentation-agno`        | `AgnoInstrumentor`       |
| AutoGen               | `openinference-instrumentation-autogen`     | `AutogenInstrumentor`    |
| Guardrails            | `openinference-instrumentation-guardrails`  | `GuardrailsInstrumentor` |

<Note>
  Two special cases:

  * **AutoGen** instruments the classic `ConversableAgent` API (the `ag2`
    package), not the newer `autogen-agentchat` rewrite.
  * **MCP** (`openinference-instrumentation-mcp` → `MCPInstrumentor`) propagates
    trace context across the MCP client/server boundary so tool calls stay on
    the same trace — it stitches spans rather than emitting model spans itself.
    Add it alongside the instrumentor for whatever runs your model.
</Note>

## Turning one on explicitly

Install the instrumentor and call it after `bryel.init()`:

```python theme={"theme":{"light":"github-light","dark":"github-dark"}}
import bryel
bryel.init(service_name="research-agent")   # exporter only

from openinference.instrumentation.crewai import CrewAIInstrumentor
CrewAIInstrumentor().instrument()
```

That's the same thing `instrument=True` does — it just calls every instrumentor
it finds. Use the explicit form when you want to trace one library and not the
rest.

<Tip>Don't see your framework? If it emits OpenTelemetry GenAI or OpenInference spans, bryel already accepts them — point your OTLP exporter at the [HTTP API](/sdk/http). If it doesn't, [tell us](https://docs.bryel.ai) and we'll add it.</Tip>