Agents Overview

Agent Integration Kit supports multiple AI agents through the Agent Client Protocol (ACP). Each agent has different capabilities, pricing, and setup requirements.

Supported Agents

How Agents Work

Built-in Agents (OpenRouter)

Built-in agents communicate via HTTP directly from the plugin:

• No external executables required
• API key stored in Project Settings
• Streaming via Server-Sent Events (SSE)

External Agents (Claude Code, Gemini CLI)

External agents run as separate processes:

• Plugin spawns the CLI process
• Communication via JSON-RPC over stdio
• Full agent capabilities including local tools

Agent Protocol (ACP)

All agents communicate using the Agent Client Protocol:

{
  "jsonrpc": "2.0",
  "method": "chat",
  "params": {
    "messages": [...],
    "tools": [...]
  }
}

The plugin handles:

• Session management
• Tool registration
• Permission requests
• Streaming responses

Choosing an Agent

OpenRouter

• Pros: Easy setup, many models, no CLI install
• Cons: Requires API key, pay-per-use

Claude Code

• Pros: Full Claude capabilities, local file access
• Cons: Requires CLI installation, Anthropic account

Gemini CLI

• Pros: Google AI integration, generous free tier
• Cons: Requires CLI installation, Google account

Multi-Agent Setup

You can configure multiple agents and switch between them:

1. Configure each agent in Project Settings
2. Save as separate profiles
3. Switch profiles in the chat window dropdown

This lets you use different agents for different tasks or compare outputs.