> ## 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.

# Anthropic

> Use Anthropic Claude models (Claude Sonnet, Claude Haiku) with Agentium. Setup, factory, and config.

# Anthropic

Use Anthropic's Claude models with Agentium through the unified `ModelProvider` interface. Claude excels at long-context tasks, analysis, and nuanced reasoning.

***

## Setup

<Tabs>
  <Tab title="Install">
    Install the Anthropic SDK (required by Agentium for Anthropic support):

    ```bash theme={null}
    npm install @anthropic-ai/sdk
    ```
  </Tab>

  <Tab title="Environment">
    Set your API key via environment variable:

    ```bash theme={null}
    export ANTHROPIC_API_KEY="sk-ant-..."
    ```

    Or pass it in config (see below).
  </Tab>
</Tabs>

***

## Factory

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

const model = anthropic("claude-sonnet-4-20250514");
```

<ParamField path="modelId" type="string" required>
  The Anthropic model identifier.
</ParamField>

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

***

## Supported Models

| Model ID                    | Description                                              |
| --------------------------- | -------------------------------------------------------- |
| `claude-sonnet-4-20250514`  | Claude Sonnet 4. Strong balance of speed and capability. |
| `claude-3-5-haiku-20241022` | Claude 3.5 Haiku. Fast, cost-effective.                  |

<Accordion title="Using a different Claude model">
  Pass any valid Anthropic model ID to the factory. Check the [Anthropic docs](https://docs.anthropic.com) for the latest model IDs.
</Accordion>

***

## Config

<ParamField path="apiKey" type="string" required={false}>
  Anthropic API key. If omitted, uses `ANTHROPIC_API_KEY` environment variable.
</ParamField>

### Example

```typescript theme={null}
const model = anthropic("claude-sonnet-4-20250514", {
  apiKey: process.env.ANTHROPIC_API_KEY,
});
```

***

## Multi-Modal Support

Anthropic Claude supports **images** and **files/documents** as input. Audio is not supported.

### Images

Pass images as base64 or URL in `ContentPart[]`:

```typescript theme={null}
const result = await agent.run([
  { type: "text", text: "What's in this image?" },
  { type: "image", data: "https://example.com/photo.jpg", mimeType: "image/jpeg" },
]);
```

### Files & Documents

PDF and other documents can be sent via URL or base64. Anthropic processes them natively as document blocks:

```typescript theme={null}
const result = await agent.run([
  { type: "text", text: "Summarize this PDF." },
  { type: "file", data: "https://example.com/report.pdf", mimeType: "application/pdf", filename: "report.pdf" },
]);
```

Both URL-based and base64-encoded files are supported. The provider auto-detects the format.

### Unsupported: Audio

Audio input is not supported by Anthropic. If passed, the provider logs a warning and substitutes a placeholder text block.

***

## Reasoning / Extended Thinking

Claude supports extended thinking — the model "thinks" step-by-step before responding. Enable it via the `reasoning` config on your agent:

```typescript theme={null}
import { Agent, anthropic } from "@agentium/core";

const agent = new Agent({
  name: "deep-thinker",
  model: anthropic("claude-sonnet-4-6"),
  instructions: "Solve problems step by step.",
  reasoning: {
    enabled: true,
    budgetTokens: 4000, // Max tokens for thinking (not shown to user)
  },
});

const result = await agent.run(
  "A farmer has 17 sheep. All but 9 run away. How many does he have left?"
);

console.log(result.thinking);
// "Let me parse this carefully... 'all but 9 run away' means 9 remain..."
console.log(result.text);
// "The farmer has 9 sheep left."
```

<Warning>
  When `reasoning` is enabled, `maxTokens` is automatically adjusted. If your `maxTokens` is lower than `budgetTokens + 4096`, it's overridden to ensure room for both thinking and response. See [Reasoning](/agents/reasoning) for details.
</Warning>

***

## Prompt Caching

Anthropic automatically caches prompt prefixes for repeated conversations. This reduces cost and latency when the system prompt or conversation history is reused across requests. No configuration is needed — the SDK handles this transparently.

Benefits are most significant when:

* System prompts are long (e.g., detailed instructions, few-shot examples)
* Conversations span many turns (the prefix is cached across requests in the same session)
* Multiple users share the same agent instructions

***

## Full Example

```typescript theme={null}
import { Agent, anthropic } from "@agentium/core";

const agent = new Agent({
  name: "Claude Assistant",
  model: anthropic("claude-sonnet-4-20250514", {
    apiKey: process.env.ANTHROPIC_API_KEY,
  }),
  instructions: "You are a thoughtful, thorough assistant. Think step by step.",
});

const output = await agent.run("Analyze the pros and cons of microservices.");
console.log(output.text);
```