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

# AWS Bedrock

> Use AWS Bedrock foundation models (Mistral, Amazon Nova, Llama, Cohere) with Agentium. Setup, factory, config, and multi-model support.

# AWS Bedrock

Use AWS Bedrock to access a broad catalog of foundation models — Mistral, Amazon Nova, Meta Llama, Cohere, AI21, and more — through Agentium's unified `ModelProvider` interface. All models are accessed via the Bedrock Converse API.

<Info>
  For **Claude models on Bedrock**, use the dedicated [`awsClaude()`](/models/aws-claude) provider instead. It uses the native Anthropic SDK for full Claude feature support (extended thinking, document input, etc.).
</Info>

***

## Setup

<Tabs>
  <Tab title="Install">
    Install the AWS Bedrock Runtime SDK:

    ```bash theme={null}
    npm install @aws-sdk/client-bedrock-runtime
    ```
  </Tab>

  <Tab title="Environment">
    Set your AWS credentials via environment variables:

    ```bash theme={null}
    export AWS_ACCESS_KEY_ID="AKIA..."
    export AWS_SECRET_ACCESS_KEY="..."
    export AWS_REGION="us-east-1"
    ```

    Or use AWS default credential chain (SSO, IAM roles, EC2 instance profiles).
  </Tab>
</Tabs>

***

## Factory

```typescript theme={null}
import { awsBedrock } from "@agentium/core";

const model = awsBedrock("mistral.mistral-large-2402-v1:0");
```

<ParamField path="modelId" type="string" required>
  The Bedrock model identifier. Find model IDs in the [AWS Bedrock model catalog](https://docs.aws.amazon.com/bedrock/latest/userguide/models-supported.html).
</ParamField>

<ParamField path="config" type="AwsBedrockConfig" required={false}>
  Optional configuration. See Config below.
</ParamField>

***

## Supported Models

| Model ID                          | Provider | Description                        |
| --------------------------------- | -------- | ---------------------------------- |
| `mistral.mistral-large-2402-v1:0` | Mistral  | Strong general-purpose performance |
| `mistral.mistral-small-2402-v1:0` | Mistral  | Fast, cost-effective               |
| `amazon.nova-pro-v1:0`            | Amazon   | General-purpose Nova model         |
| `amazon.nova-lite-v1:0`           | Amazon   | Lightweight, low-latency           |
| `amazon.nova-micro-v1:0`          | Amazon   | Ultra-fast, cost-optimized         |
| `meta.llama3-1-70b-instruct-v1:0` | Meta     | Llama 3.1 70B Instruct             |
| `meta.llama3-1-8b-instruct-v1:0`  | Meta     | Llama 3.1 8B Instruct              |
| `cohere.command-r-plus-v1:0`      | Cohere   | Strong RAG and tool use            |
| `cohere.command-r-v1:0`           | Cohere   | Fast, efficient                    |

<Accordion title="Using other Bedrock models">
  Pass any valid Bedrock model ID to the factory. Manage model access in the [AWS Bedrock console](https://us-east-1.console.aws.amazon.com/bedrock/home#/model-catalog). Not all models support all features — check [supported features per model](https://docs.aws.amazon.com/bedrock/latest/userguide/conversation-inference-supported-models-features.html).
</Accordion>

***

## Config

<ParamField path="accessKeyId" type="string" required={false}>
  AWS access key ID. Falls back to `AWS_ACCESS_KEY_ID` env var, then default credential chain.
</ParamField>

<ParamField path="secretAccessKey" type="string" required={false}>
  AWS secret access key. Falls back to `AWS_SECRET_ACCESS_KEY` env var.
</ParamField>

<ParamField path="region" type="string" required={false} default="us-east-1">
  AWS region. Falls back to `AWS_REGION` env var.
</ParamField>

<ParamField path="sessionToken" type="string" required={false}>
  AWS session token for temporary credentials (STS, SSO). Falls back to `AWS_SESSION_TOKEN` env var.
</ParamField>

***

## Authentication Methods

### Method 1: Access Key + Secret (Explicit)

```typescript theme={null}
const model = awsBedrock("mistral.mistral-large-2402-v1:0", {
  accessKeyId: "AKIA...",
  secretAccessKey: "...",
  region: "us-east-1",
});
```

### Method 2: Environment Variables (Recommended)

```bash theme={null}
export AWS_ACCESS_KEY_ID="AKIA..."
export AWS_SECRET_ACCESS_KEY="..."
export AWS_REGION="us-east-1"
```

```typescript theme={null}
const model = awsBedrock("mistral.mistral-large-2402-v1:0");
// Credentials picked up from env automatically
```

### Method 3: Default Credential Chain (SSO, IAM Roles)

If running on EC2, ECS, Lambda, or using `aws sso login`, credentials are resolved automatically via the AWS SDK's default credential provider chain:

```typescript theme={null}
const model = awsBedrock("amazon.nova-pro-v1:0", {
  region: "us-east-1",
});
// Uses IAM role, SSO session, or ~/.aws/credentials
```

### Method 4: Temporary Credentials (STS)

```typescript theme={null}
const model = awsBedrock("mistral.mistral-large-2402-v1:0", {
  accessKeyId: "ASIA...",
  secretAccessKey: "...",
  sessionToken: "FwoGZX...",
  region: "us-east-1",
});
```

***

## Tool Calling

Tool calling is supported via the Bedrock Converse API. Define tools with `defineTool` and they work automatically:

```typescript theme={null}
import { Agent, awsBedrock, defineTool } from "@agentium/core";
import { z } from "zod";

const agent = new Agent({
  name: "bedrock-assistant",
  model: awsBedrock("mistral.mistral-large-2402-v1:0"),
  instructions: "You are a helpful assistant with tool access.",
  tools: [
    defineTool({
      name: "getWeather",
      description: "Get current weather for a city",
      parameters: z.object({ city: z.string() }),
      execute: async ({ city }) => `${city}: 24°C, Sunny`,
    }),
  ],
});

const result = await agent.run("What's the weather in Mumbai?");
console.log(result.text);
```

<Warning>
  Not all Bedrock models support tool calling. Mistral Large, Amazon Nova, Cohere Command R/R+, and Meta Llama 3.1+ support function calling. Check the [Bedrock docs](https://docs.aws.amazon.com/bedrock/latest/userguide/conversation-inference-supported-models-features.html) for your model.
</Warning>

***

## Multi-Modal Support

The Bedrock Converse API supports **images** and **documents** for models that have vision capabilities (e.g., Amazon Nova).

### Images

```typescript theme={null}
import { Agent, awsBedrock, type ContentPart } from "@agentium/core";
import { readFileSync } from "node:fs";

const agent = new Agent({
  name: "vision-agent",
  model: awsBedrock("amazon.nova-pro-v1:0"),
  instructions: "Describe images in detail.",
});

const imageData = readFileSync("photo.jpg").toString("base64");
const result = await agent.run([
  { type: "text", text: "What's in this image?" },
  { type: "image", data: imageData, mimeType: "image/jpeg" },
] as ContentPart[]);
```

### Documents

```typescript theme={null}
const pdfData = readFileSync("report.pdf").toString("base64");
const result = await agent.run([
  { type: "text", text: "Summarize the key points." },
  { type: "file", data: pdfData, mimeType: "application/pdf", filename: "report.pdf" },
] as ContentPart[]);
```

***

## Full Example

```typescript theme={null}
import { Agent, awsBedrock, CostTracker, defineTool } from "@agentium/core";
import { z } from "zod";

const costTracker = new CostTracker({
  pricing: {
    "mistral.mistral-large-2402-v1:0": { promptPer1k: 0.004, completionPer1k: 0.012 },
  },
});

const agent = new Agent({
  name: "bedrock-agent",
  model: awsBedrock("mistral.mistral-large-2402-v1:0", {
    region: "us-east-1",
  }),
  instructions: "You are a helpful assistant running on AWS Bedrock.",
  tools: [
    defineTool({
      name: "lookup_order",
      description: "Look up an order by ID",
      parameters: z.object({ orderId: z.string() }),
      execute: async ({ orderId }) => JSON.stringify({ id: orderId, status: "shipped", eta: "2026-03-01" }),
    }),
  ],
  costTracker,
  maxToolRoundtrips: 3,
});

const result = await agent.run("Where is my order ORD-12345?");
console.log(result.text);
console.log(`Cost: $${costTracker.getSummary().totalCost.toFixed(4)}`);
```

***

## AWS Bedrock vs Direct API

| Feature        | Direct API (`openai()`, `anthropic()`) | `awsBedrock()`             |
| -------------- | -------------------------------------- | -------------------------- |
| Auth           | Provider API key                       | AWS IAM / credentials      |
| Billing        | Provider billing                       | AWS billing (consolidated) |
| VPC / Private  | No                                     | Yes (PrivateLink)          |
| Model variety  | Single provider                        | Multi-provider catalog     |
| Claude support | Full (thinking, docs)                  | Use `awsClaude()` instead  |
| Compliance     | Provider-dependent                     | SOC2, HIPAA, FedRAMP       |

***

## Environment Variables

| Variable                | Description                       |
| ----------------------- | --------------------------------- |
| `AWS_ACCESS_KEY_ID`     | AWS access key ID                 |
| `AWS_SECRET_ACCESS_KEY` | AWS secret access key             |
| `AWS_REGION`            | AWS region (default: `us-east-1`) |
| `AWS_SESSION_TOKEN`     | Temporary session token (STS/SSO) |

***

## Cross-References

* [AWS Claude](/models/aws-claude) — Claude models on Bedrock with full Anthropic feature support
* [Custom Provider](/models/custom-provider) — Build your own provider for any LLM
* [Cost Tracking](/cost/overview) — Add custom pricing for Bedrock models