Anthropic¶

Autobot supports Anthropic as an LLM provider via the Messages API. This gives access to the Claude model family including Claude Opus, Sonnet, and Haiku.

Setup¶

1. Get an API key¶

Create an API key at console.anthropic.com.

2. Configure credentials¶

Add your API key to the .env file:

ANTHROPIC_API_KEY=sk-ant-...

Or use the interactive setup:

autobot setup
# Select "Anthropic (Claude)" as provider

3. Configure the provider¶

In config.yml:

agents:
  defaults:
    model: "anthropic/claude-haiku-4-5"

providers:
  anthropic:
    api_key: "${ANTHROPIC_API_KEY}"

4. Verify¶

autobot doctor
# Should show: ✓ LLM provider configured (anthropic)

Model naming¶

Models use the anthropic/ prefix followed by the Anthropic model ID:

# Latest (recommended)
model: "anthropic/claude-haiku-4-5"     # Fast, cheap, capable (recommended default)
model: "anthropic/claude-sonnet-4-6"    # Mid-tier — strong coding and reasoning
model: "anthropic/claude-opus-4-6"      # Flagship — smartest, 1M context

# Previous generation (still available)
model: "anthropic/claude-sonnet-4-5"
model: "anthropic/claude-opus-4-5"
model: "anthropic/claude-sonnet-4"
model: "anthropic/claude-opus-4"

claude-haiku-4-5 is a good default for most use cases — fast, affordable ($1/$5 per 1M tokens), and capable enough for tool use, coding, and everyday tasks. Use Sonnet or Opus when you need stronger reasoning or larger context.

The anthropic/ prefix tells autobot to route to the Anthropic Messages API. It is stripped before sending to the API.

See the full model list in the Anthropic docs.

Custom API base¶

To use a self-hosted proxy or alternative endpoint:

providers:
  anthropic:
    api_key: "${ANTHROPIC_API_KEY}"
    api_base: "https://your-proxy.example.com/v1/messages"

Extra headers¶

Pass additional headers with every request (e.g. for proxy authentication):

providers:
  anthropic:
    api_key: "${ANTHROPIC_API_KEY}"
    extra_headers:
      X-Custom-Header: "value"

Configuration reference¶

Field	Required	Default	Description
`api_key`	Yes	—	Anthropic API key (`sk-ant-...`)
`api_base`	No	`https://api.anthropic.com/v1/messages`	Custom API endpoint
`extra_headers`	No	—	Additional HTTP headers for every request

How it works¶

Anthropic uses its own Messages API format, which differs from the OpenAI-compatible standard:

x-api-key header for authentication (not Authorization: Bearer)
anthropic-version header sent with every request (2023-06-01)
System prompt extracted from messages and sent as a top-level system field
Tool use blocks use tool_use / tool_result format instead of function

Autobot detects Anthropic models automatically and handles all format conversion transparently. Tools, MCP servers, plugins, and all other features work the same as with other providers.

Voice transcription¶

Anthropic does not provide a transcription API. If you need voice message support, configure an additional Groq or OpenAI provider for Whisper-based transcription.

Known limitations¶

No streaming — Responses are returned in full after the model finishes generating.
Text and image content only — Document and audio content blocks are not supported.
Tool choice is always auto — There is no configuration to force a specific tool or disable tool use per-request.

Troubleshooting¶

Enable debug logging to see request/response details:

LOG_LEVEL=DEBUG autobot agent -m "Hello"

Look for:

POST https://api.anthropic.com/v1/messages model=... (anthropic) — confirms native Anthropic path
Response 200 (N bytes) — confirms API response
HTTP 4xx/5xx: ... — API errors with details

Common issues¶

"No LLM provider configured" — Check that api_key is set and non-empty in config.yml.

"API error: authentication_error" — Invalid or expired API key. Verify at console.anthropic.com.

"API error: rate_limit_error" — Too many requests. Anthropic applies per-key rate limits — check your plan's limits.

"API error: overloaded_error" — The API is temporarily overloaded. Retry after a moment.