Anthropic Claude

Install the optional SDK extra:

python -m pip install "agent-company-core[anthropic]"
$env:ANTHROPIC_API_KEY = "..."

The adapter uses the official async Anthropic Messages API and the standard ANTHROPIC_API_KEY environment variable. The model identifier is required at construction time. Anthropic currently documents claude-sonnet-4-6 as an active example; keep it configurable and check Anthropic's model lifecycle documentation before deploying.

import asyncio
import os
from pathlib import Path

from agent_company_core import AnthropicModel, MissionCreate, MissionEngine
from agent_company_core.persistence import SQLiteStore


async def main() -> None:
    model = AnthropicModel(
        model=os.environ.get("ANTHROPIC_MODEL", "claude-sonnet-4-6")
    )
    engine = MissionEngine(
        store=SQLiteStore(Path("runtime/state.sqlite3")),
        providers={"fast": model, "reasoning": model, "vision": model},
    )
    mission = engine.create_mission(
        MissionCreate(title="Claude mission", objective="Return a harmless typed decision")
    )
    result = await engine.run_mission(mission.id)
    print(result.status.value, result.summary)


asyncio.run(main())

The framework asks the model for JSON when a response schema is supplied, then validates the result as AgentDecision. Provider failures are normalized so the engine can use its declared fallback aliases. The adapter never writes the API key to audit output or exception text.

The repository example is safe by default: it injects a fake client and makes no paid call. Use --live only in a controlled local environment with an explicitly configured key.