Guide

Build an agent, give it tools, and run it in your application. Start with Getting started for a working example, or read What is Effect Agent? for an introduction.

Build your agent

• Agent definitions: define inputs, outputs, instructions, and execution limits.
• Tools and layers: connect models to your application services.
• Run and stream: execute an agent and observe its progress.
• Threads: keep conversation history across runs.
• Context management: manage long conversations and retrieved context.

Add capabilities

• Subagents: delegate work and choose whether the parent waits or continues.
• Agent messaging: exchange input between independent agents.
• Effect Workflows: drive durable execution through a workflow engine.
• Sandbox execution: run commands in a controlled environment.
• Code Mode: let agents compose authorized tools with generated JavaScript.
• Browser tools: capture pages, crawl sites, and interact with browsers.

Test and operate

Use deterministic testing to exercise behavior without live model calls. The operations guide covers authorization, recovery, and scheduled work. Storage adapter authors can run the certification contracts.

When you are ready to host durable work, choose a platform.