LLM

The agent talks to language models through the LLMClient contract.

import { anthropic, type LLMClient } from "duclaw-cli/sdk";

LLMClient

Implement this contract when you want to connect a custom model provider.

type LLMClient = {
  chat(
    messages: LLMMessage[],
    system: string,
    tools?: Tool[],
    options?: {
      requestId?: string;
      signal?: AbortSignal;
    },
  ): Promise<LLMResponse>;
};

Anthropic Adapter

Use anthropic() when the model endpoint speaks the Anthropic Messages API.

const model = anthropic({
  apiKey: process.env.ANTHROPIC_API_KEY!,
  baseURL: process.env.ANTHROPIC_BASE_URL,
  model: process.env.ANTHROPIC_MODEL,
});

baseURL is optional. Pass it when you route requests through an Anthropic-compatible gateway, proxy, or private model endpoint:

const model = anthropic({
  apiKey: process.env.MODEL_API_KEY!,
  baseURL: "https://gateway.example.com/anthropic",
  model: "claude-sonnet-4-6",
});

If the endpoint expects bearer authentication instead of Anthropic's standard x-api-key header, set authStyle:

const model = anthropic({
  apiKey: process.env.MODEL_API_KEY!,
  baseURL: "https://gateway.example.com/anthropic",
  model: "claude-sonnet-4-6",
  authStyle: "bearer",
});

Supported options:

Option	Required	Purpose
apiKey	Yes	API key or gateway token
model	No	Model name sent to the endpoint
baseURL	No	Custom Anthropic-compatible endpoint
maxTokens	No	Maximum response tokens
authStyle	No	Use "x-api-key" by default, or "bearer" for Authorization: Bearer gateways

Use a custom LLMClient when your application owns model routing, retries, observability, or provider fallback.