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

# AWS Bedrock

> Auto-track all AWS Bedrock LLM calls

## Installation

```bash theme={null}
pip install meshai-sdk[bedrock]
```

## Usage

```python theme={null}
from meshai import MeshAI
from meshai.integrations.bedrock import wrap_bedrock
import boto3

client = MeshAI(api_key="msh_...", agent_name="my-agent")
client.register(framework="custom", model_provider="aws")

# Wrap the Bedrock client
bedrock = wrap_bedrock(
    boto3.client("bedrock-runtime"),
    meshai=client,
)

# Use converse(), auto-tracked
response = bedrock.converse(
    modelId="anthropic.claude-sonnet-4-6-20250514-v1:0",
    messages=[{"role": "user", "content": [{"text": "Hello"}]}],
)

# invoke_model() is also tracked
response = bedrock.invoke_model(
    modelId="amazon.titan-text-express-v1",
    body='{"inputText": "Hello"}',
)
```

## How It Works

`wrap_bedrock` patches both `converse()` and `invoke_model()` on the Bedrock runtime client. After each call, it:

1. Extracts the **model name** from the `modelId` parameter
2. Extracts **token counts** from the response usage metadata
3. Infers the **provider** from the model ID prefix (e.g., `anthropic.`, `amazon.`, `meta.`)
4. Sends the usage event to MeshAI (buffered, non-blocking)

## Alternative: Proxy (Zero-Code)

For Bedrock, the proxy approach requires configuring a custom endpoint in your boto3 client. The region must be part of the path:

```python theme={null}
bedrock = boto3.client(
    "bedrock-runtime",
    endpoint_url="https://proxy.meshai.dev/v1/bedrock/k/msh_YOUR_PROXY_KEY/us-east-1",
)
```