autogpt-agents

Start the AutoGPT platform with Docker Compose, open the visual builder at localhost:3000, add blocks from the BlocksControl panel, connect nodes by dragging between handles, configure inputs, and run the agent. Agents are graphs of nodes containing reusable blocks.

AutoGPT suits continuous agents with visual workflows and external triggers. LangChain offers more control over agent logic in code, CrewAI focuses on role-based multi-agent collaboration, and OpenAI Assistants fits simple hosted deployments.

Send a POST request to /api/v1/webhooks/{webhook_id} with a JSON payload. The platform verifies the signature, parses the payload, and queues the associated graph for execution, returning an execution ID.

Stuck executions usually indicate executor or queue issues. Check executor logs with docker compose logs executor, inspect the RabbitMQ management UI at localhost:15672, and cancel the execution via the API's cancel endpoint.

Yes, custom blocks subclass the Block class with defined input and output schemas, an async execute method that yields outputs, and optional credential requirements. Register blocks in the backend blocks module to make them available in the UI.

Avoid AutoGPT when you need fine-grained programmatic control over agent logic, lightweight single-run scripts, or deep Microsoft ecosystem integration. LangChain, Semantic Kernel, or simple hosted assistants fit those cases better.