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

# PydanticAI

> Trace every PydanticAI agent run with one function.

## Install

<CodeGroup>
  ```bash Python theme={null}
  pip install trulayer pydantic-ai
  ```

  ```bash TypeScript theme={null}
  npm install @trulayer/sdk pydantic-ai
  ```
</CodeGroup>

## Instrument

<CodeGroup>
  ```python Python theme={null}
  import os
  import trulayer

  trulayer.init(api_key=os.environ["TRULAYER_API_KEY"], project_name="my-app")
  trulayer.instrument_pydanticai()
  ```

  ```typescript TypeScript theme={null}
  import { TruLayer, instrumentPydanticai } from "@trulayer/sdk";

  const tl = new TruLayer({
    apiKey: process.env.TRULAYER_API_KEY!,
    projectName: "my-app",
  });

  instrumentPydanticai(tl);
  ```
</CodeGroup>

After instrumentation every `Agent.run` / `Agent.run_sync` call and every tool invocation is captured — no per-callsite changes needed.

## Minimal example

<CodeGroup>
  ```python Python theme={null}
  from pydantic_ai import Agent

  agent = Agent("openai:gpt-4o-mini", system_prompt="You are a concise assistant.")

  result = agent.run_sync("What is the capital of France?")
  print(result.output)
  # A trace appears in the dashboard with one `agent` span and one `llm` child.
  ```

  ```typescript TypeScript theme={null}
  import { Agent } from "pydantic-ai";

  const agent = new Agent({
    model: "openai:gpt-4o-mini",
    systemPrompt: "You are a concise assistant.",
  });

  const result = await agent.run("What is the capital of France?");
  console.log(result.output);
  ```
</CodeGroup>

## What gets captured

* `agent` spans for every `run` / `run_sync` / `run_stream` call with the user prompt, system prompt, and final structured output
* `llm` spans for each underlying model call, including token counts and any tool-call requests from the model
* `tool` spans for each `@agent.tool` invocation with validated arguments and return value
* Validation errors from Pydantic are captured on the span as errors so you can filter failed runs in the dashboard