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

# Supported frameworks

> BotGauge verifies tracing for 11 AI agent frameworks, including LangChain, OpenAI, Anthropic, CrewAI, Google ADK, and MCP. See extras and verified versions.

Each extra installs the [OpenInference](https://github.com/Arize-ai/openinference) instrumentor for one framework. `botgauge instrument` traces that framework at whatever version you have installed. The versions below are the ones BotGauge verified.

## Verified versions

| Framework | Extra | Verified versions | Zero-code run |
| - | - | - | - |
| [LangChain](https://www.botgauge.com/docs/integrations/langchain.md) | `[langchain]` | langchain 1.3.15, 1.4.6 with `openinference-instrumentation-langchain` 0.1.79. Also langchain 1.3.15 with 0.1.76. | Traced |
| [LangGraph](https://www.botgauge.com/docs/integrations/langgraph.md) | `[langgraph]` | langgraph 1.2.11, 1.2.16 with `openinference-instrumentation-langchain` 0.1.79. Also langgraph 1.2.11 with 0.1.76. | Traced |
| [OpenAI SDK](https://www.botgauge.com/docs/integrations/openai.md) | `[openai]` | openai 2.45.0, 2.54.0, 3.33.0, 3.35.1 with `openinference-instrumentation-openai` 0.1.65. Also openai 2.45.0 with 0.1.61. | Traced |
| [Anthropic SDK](https://www.botgauge.com/docs/integrations/anthropic.md) | `[anthropic]` | anthropic 1.8.0, 1.15.0 with `openinference-instrumentation-anthropic` 3.0.3. Also anthropic 1.8.0 with 2.1.6. | Traced |
| [LlamaIndex](https://www.botgauge.com/docs/integrations/llamaindex.md) | `[llamaindex]` | llama-index-core 0.14.25 with `openinference-instrumentation-llama-index` 4.6.0. Also 0.14.25 with 4.5.2. | Traced |
| [CrewAI](https://www.botgauge.com/docs/integrations/crewai.md) | `[crewai]` | crewai 1.15.22, 1.15.28 with `openinference-instrumentation-crewai` 1.1.21. Also crewai 1.15.22 with 1.1.18. Plus the OpenAI producer: openai 2.45.0, 2.54.0 with `openinference-instrumentation-openai` 0.1.65. | Traced |
| [OpenAI Agents](https://www.botgauge.com/docs/integrations/openai-agents.md) | `[openai-agents]` | openai-agents 0.20.0, 0.24.0 with `openinference-instrumentation-openai-agents` 2.5.3. Also openai-agents 0.20.0 with 2.5.0. | Traced |
| [Google ADK](https://www.botgauge.com/docs/integrations/google-adk.md) | `[google-adk]` | google-adk 2.10.0, 2.12.0 with `openinference-instrumentation-google-adk` 1.0.4. Also google-adk 2.10.0 with 1.0.0. | Traced (model through LiteLLM). Needs google-adk 2.10 or newer. |
| [AutoGen AgentChat](https://www.botgauge.com/docs/integrations/autogen.md) | `[autogen-agentchat]` | autogen-agentchat 0.7.5 with `openinference-instrumentation-autogen-agentchat` 0.1.22. Also 0.7.5 with 0.1.18. | Traced |
| [PydanticAI](https://www.botgauge.com/docs/integrations/pydantic-ai.md) | `[pydantic-ai]` | pydantic-ai 1.25.1, 2.57.0 with `openinference-instrumentation-pydantic-ai` 0.1.30. Also pydantic-ai 1.25.1 with 0.1.27. | Traced |
| [MCP](https://www.botgauge.com/docs/integrations/mcp.md) | `[mcp]` | mcp 1.24.0, 2.3.0 with `openinference-instrumentation-mcp` 2.0.13. Also mcp 1.24.0 with 2.0.10. | Traced (an LLM calls a tool on an in-process MCP server, with `[openai]` for the LLM) |

```bash terminal theme={null}
# the extra, plus your framework (a verified version shown, or your own)
pip install "botgauge[langgraph]" langgraph==1.2.16 langchain-openai==1.7.2
```

## What "verified" means

A framework version is verified only together with the instrumentor version listed under it, which is the one the extra installs, and only the exact versions listed are verified. The oldest and newest version of each framework are certified on every change to BotGauge, and every other listed pair is certified every night. The word "Also" names an older instrumentor version that is verified with one framework version. The extra no longer installs it.

In the last column, **Traced** means the unmodified app, launched with `botgauge instrument`, exported the framework's spans with the `turn()` session id, masked the email address and card number in the prompt, and, for the frameworks that call an LLM, exported the LLM call with its model and input and output token counts. Cost is computed from those token counts. Every sample calls a real LLM.

## Notes on specific frameworks

CrewAI calls its model through the OpenAI SDK, which the CrewAI instrumentor doesn't trace, so the `[crewai]` extra also installs the OpenAI producer. Google ADK reaches the model through its LiteLLM wrapper. In the MCP sample, the LLM calls a tool on an in-process MCP server.

Each framework was installed with exactly the install line on its page. That resolved OpenTelemetry 1.45.1 for most frameworks. Google ADK needs OpenTelemetry 1.42.1 or older, which `pip` picks for you.

PydanticAI 1.25, the older verified version, needs OpenTelemetry 1.39.1 and `genai-prices` 0.0.73. PydanticAI reads token counts through `genai-prices`, and every PydanticAI before 2.17 records none with `genai-prices` 0.1.0 or newer, so the trace shows zero cost. PydanticAI 2.17 or newer needs neither pin. In code, pass `version=3` to `InstrumentationSettings` on PydanticAI 1.x and `version=5` on 2.x. `botgauge instrument` picks it for you.

Start tracing your AI agentsExplore every supported framework and start sending traces to [BotGauge](https://cloud.botgauge.com).

[Try BotGauge →](https://cloud.botgauge.com)

## Running a version that isn't listed

Other versions may work but aren't verified. See [versions not verified](https://www.botgauge.com/docs/integrations/versions-not-verified.md) for what that means and how to enforce verified versions in CI.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.