VS Code Chat Participant Integration

The Hanzo extension now includes full VS Code Chat Participant integration, making all 53+ MCP tools available through VS Code's chat interface.

Features

1. Chat Participant Registration

Registered as @hanzo (lowercase) in VS Code's chat interface
Accessible via Ctrl+Alt+I (Cmd+Alt+I on Mac)
Shows up with the Hanzo icon

2. LLM Integration

The chat participant supports multiple LLM providers:

Hanzo AI Gateway (default): Uses api.hanzo.ai
LM Studio: Local models via http://localhost:1234
Ollama: Local models via http://localhost:11434
OpenAI: Direct OpenAI API integration
Anthropic: Direct Claude API integration

3. Tool Access

All MCP tools are available through natural language or direct commands:

File Operations:

@hanzo read src/index.ts
@hanzo write config.json content: { "key": "value" }
@hanzo edit file.ts replace "old" with "new"

Search Operations:

@hanzo search "TODO"
@hanzo grep "function" in **/*.ts

Shell Operations:

@hanzo run npm test
@hanzo bash git status

AI Operations:

@hanzo agent analyze this codebase for security issues
@hanzo dispatch_agent implement new feature X

4. Configuration

Add to VS Code settings.json:

{
  // Choose LLM provider
  "hanzo.llm.provider": "hanzo", // or "lmstudio", "ollama", "openai", "anthropic"
  
  // Hanzo AI Gateway (default)
  "hanzo.llm.hanzo.apiKey": "your-api-key",
  
  // LM Studio (local)
  "hanzo.llm.lmstudio.endpoint": "http://localhost:1234/v1",
  "hanzo.llm.lmstudio.model": "model-name",
  
  // Ollama (local)
  "hanzo.llm.ollama.endpoint": "http://localhost:11434",
  "hanzo.llm.ollama.model": "llama2",
  
  // OpenAI
  "hanzo.llm.openai.apiKey": "sk-...",
  "hanzo.llm.openai.model": "gpt-4",
  
  // Anthropic
  "hanzo.llm.anthropic.apiKey": "sk-ant-...",
  "hanzo.llm.anthropic.model": "claude-3-opus-20240229"
}

5. Agent Tool

The agent and dispatch_agent tools now support all configured LLM providers:

// Single agent
@hanzo agent analyze the security of this codebase

// Multi-agent dispatch
@hanzo dispatch_agent implement user authentication with JWT

The agent tool will:

Use the configured LLM provider
Have access to specified MCP tools
Execute tasks with proper context
Support both sequential and parallel execution

6. Privacy with Local LLMs

For complete privacy, use LM Studio or Ollama:

LM Studio: Download from lmstudio.ai, load a model, start the server
Ollama: Install via brew install ollama, run ollama pull llama2

Then set "hanzo.llm.provider": "lmstudio" or "ollama" in settings.

Installation

Install the VSIX: code --install-extension dist/hanzoai-1.5.4.vsix
Configure your preferred LLM provider in settings
Open VS Code chat (Ctrl+Alt+I)
Type @hanzo to start using the assistant

Usage Examples

@hanzo show available tools
@hanzo read package.json
@hanzo search for TODO comments
@hanzo run the tests
@hanzo agent refactor this function to be more efficient

The assistant will use the configured LLM to understand your intent and execute the appropriate tools.

3.2 KiB Raw Permalink Blame History