Phoenix Agent Documentation

Everything you need to know about installing, configuring, and extending Phoenix Agent.

What is Phoenix Agent?

Phoenix Agent is an autonomous AI system agent for Windows. Unlike chatbots that only talk, Phoenix Agent acts: it executes shell commands, opens applications, controls your mouse and keyboard, manages files, and automates complex workflows — all driven by natural language.

Key principle: Phoenix Agent is an agent, not a chatbot. When you ask it to do something, it executes the action directly on your system.

Quick Start

1. Install

npm install -g phoenix-agent

2. Launch

phoenix

3. Connect a provider

/connect

Follow the interactive wizard to select your AI provider and enter your API key. Keys are encrypted locally with AES-128.

4. Action authorization

/yolo on

By default, every shell, mouse, or keyboard action asks for your confirmation before running. Enable yolo mode to let the agent act autonomously. All actions are logged either way.

5. Start controlling

ouvre le navigateur
crée un dossier projets sur le bureau
installe node.js

Core Concepts

Concept	Description
`Authorization`	By default, every shell, mouse, or keyboard action requires your confirmation before executing.
`Yolo Mode`	When enabled (`/yolo on`), actions execute without per-step confirmation. All operations are still logged.
`[EXEC] blocks`	The AI wraps shell commands in `[EXEC]...[/EXEC]` tags. Phoenix extracts and executes them.
`[MOUSE] blocks`	Mouse actions (move, click, drag, scroll) via `[MOUSE]...[/MOUSE]` tags.
`[KEYS] blocks`	Keyboard actions (type, hotkey, press) via `[KEYS]...[/KEYS]` tags.
`Decision Journal`	Every significant decision is recorded with reasoning for audit and debugging.
`Memory`	Persistent conversation history and knowledge base with full-text search.

Supported Providers

Provider	Type	API Key
OpenAI	Cloud	Required
Anthropic (Claude)	Cloud	Required
Google (Gemini)	Cloud	Required
xAI (Grok)	Cloud	Required
DeepSeek	Cloud	Required
Mistral	Cloud	Required
OpenRouter	Cloud (100+ models)	Required
Ollama	Local	None
LM Studio	Local	None
Azure OpenAI	Enterprise	Required

Next Steps

User Guide — Complete walkthrough of all features
Architecture — How Phoenix Agent is built
API Reference — Programmatic usage
Plugin Development — Extend Phoenix with custom commands
Security Model — Permissions, encryption, audit