Agent
Overview

Agent

The Agent is ProxyAI's autonomous mode for multi-step work. It plans and executes tasks, calls tools on your behalf, and can delegate specific work to subagents when it needs focused help.

What it can do

  • Plan and execute multi-step work with approvals for risky actions
  • Support multi-turn conversations by asking clarifying questions when needed
  • Read, search, edit, and run commands across your project
  • Delegate focused work to Subagents
  • Load reusable instruction packs with Skills
  • Extend and control runs with Hooks (for example: block risky operations, log tool usage, build usage dashboards)
  • Follow project-specific instructions from a PROXYAI.md file at the repo root
  • Run with many different models and providers (see Providers)
  • Automatically compress long-running context to stay within model limits

Where to go next

  • Learn how specialized agents work in Subagents
  • Define reusable instruction packs in Skills
  • Configure guardrails in Security
  • Customize automation with Hooks
  • See the available tools in Tools