DuckAI (duck.ai)¶

Autobot supports DuckDuckGo AI Chat as an LLM provider via the DuckAI proxy. This gives access to multiple models for free, with no API key required.

How it works¶

DuckDuckGo offers free AI chat at duck.ai. The API uses a custom authentication mechanism that requires a JavaScript runtime, so a lightweight proxy translates it into an OpenAI-compatible API.

The DuckAI proxy runs locally and handles this translation. Autobot connects to it like any other OpenAI-compatible endpoint.

Available models¶

Model	Description
`gpt-4o-mini`	OpenAI GPT-4o Mini (recommended)
`claude-3-5-haiku-latest`	Anthropic Claude 3.5 Haiku
`meta-llama/Llama-4-Scout-17B-16E-Instruct`	Meta Llama 4 Scout
`mistralai/Mistral-Small-24B-Instruct-2501`	Mistral Small

Model availability depends on DuckDuckGo and may change at any time. The proxy may list models that are temporarily unavailable upstream. Check working models via:

curl http://localhost:3000/v1/models

Setup¶

1. Start the DuckAI proxy¶

Using Docker:

docker run -d -p 3000:3000 --name duckai docker.io/amirkabiri/duckai

Using Podman:

podman run -d -p 3000:3000 --name duckai docker.io/amirkabiri/duckai

Verify it's running:

curl http://localhost:3000/health

2. Configure the provider¶

Add the proxy URL to your .env file:

DUCKAI_API_BASE=http://localhost:3000/v1

In config.yml:

agents:
  defaults:
    model: "duckai/gpt-4o-mini"

providers:
  duckai:
    api_base: "${DUCKAI_API_BASE}"
    api_key: "unused"

No API key is needed. DuckAI is a free service with no authentication. The api_key field is required by the config schema but the proxy ignores it — set it to any non-empty value like "unused".

3. Verify¶

autobot doctor
# Should show: LLM provider configured (duckai)

Model naming¶

Models use the duckai/ prefix followed by the model ID:

model: "duckai/gpt-4o-mini"
model: "duckai/claude-3-5-haiku-latest"
model: "duckai/meta-llama/Llama-4-Scout-17B-16E-Instruct"
model: "duckai/mistralai/Mistral-Small-24B-Instruct-2501"

The duckai/ prefix tells autobot to route to the DuckAI proxy. It is stripped before sending to the API.

Configuration reference¶

Field	Required	Default	Description
`api_base`	Yes	-	DuckAI proxy URL (e.g. `${DUCKAI_API_BASE}`)
`api_key`	Yes*	-	Any non-empty value — no real key needed, proxy ignores it
`extra_headers`	No	-	Additional HTTP headers for every request

*Required by config schema, but the value does not matter. DuckAI is free with no authentication.

Rate limits¶

DuckDuckGo enforces rate limits on the AI chat service:

~20 requests per 60-second window
Minimum 1 second between requests
The proxy handles backoff automatically

For heavy usage, consider using a paid provider instead.

Known limitations¶

No system role - DuckDuckGo's API does not support the system message role. Autobot automatically merges system prompts into the first user message.
Rate limited - Free service with usage limits. Not suitable for production workloads.
Proxy required - The DuckAI proxy must be running alongside autobot.
Model availability - Models may be listed by the proxy but rejected by DuckDuckGo's backend. Availability changes without notice.
No voice transcription - If you need voice message support, configure an additional Groq or OpenAI provider.

Troubleshooting¶

Enable debug logging to see request/response details:

LOG_LEVEL=DEBUG autobot agent -m "Hello"

Common issues¶

"No LLM provider configured" - Check that api_key is set (any non-empty value) and api_base points to the running proxy.

"LLM request failed: Connection refused" - The DuckAI proxy is not running. Start it with docker run -p 3000:3000 docker.io/amirkabiri/duckai.

"API error: DuckAI API error: 400 Bad Request" - The model is not available. Some models may be listed by the proxy but rejected by DuckDuckGo's backend. Try a different model (e.g. gpt-4o-mini).

"API error: 429 Too Many Requests" - Rate limited. Wait a moment and try again. The proxy handles backoff automatically for subsequent requests.