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

# Authentication

> Configure AI provider API keys for cloud inference

agtOS uses API keys to authenticate with cloud AI providers. You can configure one or both providers — the system uses whichever is configured in the model slot registry.

## Anthropic (Claude)

Pay-per-token billing from [console.anthropic.com](https://console.anthropic.com).

```bash theme={null}
# .env.local
ANTHROPIC_API_KEY=sk-ant-api03-your-key-here
```

* Token prefix: `sk-ant-api03-`
* Sent via `X-Api-Key` HTTP header
* Direct SDK connection to `api.anthropic.com`

## OpenAI

Pay-per-token billing from [platform.openai.com](https://platform.openai.com).

```bash theme={null}
# .env.local
OPENAI_API_KEY=sk-your-key-here
```

* Token prefix: `sk-`
* Used via OpenAI Node SDK v6
* Supports GPT-4o and GPT-4o Mini

## Cloud Provider Selection

Cloud provider selection is configured per slot in `~/.agtos/config.json` via the Model Slot Registry ([ADR-020](/architecture/decisions#adr-020)). Run `agtos setup` to configure slots interactively, or edit the config file directly:

```json ~/.agtos/config.json theme={null}
{
  "slots": {
    "chat": { "provider": "claude", "model": "claude-sonnet-4-20250514" },
    "reasoning": { "provider": "openai", "model": "gpt-4o" }
  }
}
```

Each slot can use a different provider (`claude`, `openai`, `ollama`, `openrouter`). When both API keys are configured, you can mix providers across slots — e.g., Claude for chat and OpenAI for reasoning.

## Claude Transport Options

By default, agtOS connects to the Anthropic API directly via the SDK. You can optionally spawn the `claude` CLI binary as a subprocess:

```bash theme={null}
# .env.local
AGTOS_CLAUDE_TRANSPORT=sdk   # default — direct API via Anthropic SDK
# AGTOS_CLAUDE_TRANSPORT=cli # alternative — spawn claude CLI binary
```

When using CLI transport, the `ANTHROPIC_API_KEY` is passed to the subprocess as an environment variable.

<Note>
  CLI transport requires the `claude` CLI to be installed and on your PATH. Install from [claude.ai/download](https://claude.ai/download).
</Note>

## Advanced Options

```bash theme={null}
# Adaptive thinking (recommended for Claude 4.6)
CLAUDE_THINKING=adaptive
CLAUDE_EFFORT=medium      # low, medium, high, max

# Service tier
CLAUDE_SERVICE_TIER=auto  # auto or standard_only

# Custom API endpoint (for proxies)
ANTHROPIC_BASE_URL=http://localhost:8317
```

## Endpoint Authentication

Optionally protect all `/api/*` routes with a Bearer token:

```bash theme={null}
# .env.local
AGTOS_API_KEY=your-secret-api-key
```

When set, all API requests must include `Authorization: Bearer <key>`. Uses timing-safe SHA-256 comparison. When unset, endpoints are open (suitable for local development).

## What's next

<CardGroup cols={2}>
  <Card title="Providers" icon="brain" href="/configuration/providers">
    Configure Claude, OpenAI, Ollama, sherpa-onnx, and the model router.
  </Card>

  <Card title="Environment Variables" icon="list" href="/configuration/environment">
    Complete reference for all configuration options.
  </Card>
</CardGroup>