glee 0.3.5

The Conductor for Your AI Orchestra

Glee

The Stage Manager for Your AI Orchestra

An orchestration layer for AI coding agents with shared memory, code review, and subagent orchestration.

Quick Start

# Install
uv tool install glee --python 3.13
# or: pipx install glee

# Initialize project (registers MCP server for Claude Code)
glee init claude                  # Use 'claude', 'codex', 'gemini', 'cursor', etc.

# Configure reviewers
glee config set reviewer.primary codex
glee config set reviewer.secondary gemini

# View status
glee status

# Run review
glee review src/main.py
glee review git:changes          # Review uncommitted changes
glee review git:staged           # Review staged changes

Features

• MCP Integration: glee init registers Glee as an MCP server - Claude Code gets glee_* tools automatically
• Structured Reviews: Severity levels (MUST/SHOULD, HIGH/MEDIUM/LOW) for prioritized feedback
• Reviewer Preferences: Primary + optional secondary reviewer
• Persistent Memory: Project context persists across sessions
• Stream Logging: All agent output logged to .glee/stream_logs/

Claude Code Integration

After running glee init, restart Claude Code. You'll have these MCP tools:

• glee_status - Show project status and reviewer config
• glee_review - Run code review with primary reviewer
• glee_config_set - Set config value (e.g., reviewer.primary)
• glee_config_unset - Unset config value (e.g., reviewer.secondary)
• glee_memory_add - Add a memory entry to a category
• glee_memory_list - List memories, optionally filtered by category
• glee_memory_delete - Delete memory by ID or category
• glee_memory_search - Semantic search across memories
• glee_memory_overview - Memory overview for context
• glee_memory_stats - Memory statistics
• glee_memory_bootstrap - Bootstrap memory from docs + structure

Session Hooks (automatic):

• SessionStart → injects warmup context
• SessionEnd → uses LLM to generate structured summary (goal, decisions, open_loops) and saves to memory

# In Claude Code, you can now say:
"Use glee_review to review the uncommitted changes"
"Set codex as my primary reviewer using glee"

CLI Commands

glee init <agent>                 # Initialize project + register MCP server
glee status                       # Show project status

# Configuration
glee config get                   # Show all config
glee config set reviewer.primary codex
glee config set reviewer.secondary gemini
glee config unset reviewer.secondary

# Review
glee review src/api/              # Review a directory
glee review src/main.py           # Review a file
glee review git:changes           # Review uncommitted changes
glee review git:staged            # Review staged changes

# Agents
glee test-agent codex             # Test an agent
glee mcp                          # Run MCP server (used by Claude Code)

Configuration

# .glee/config.yml
project:
  id: 550e8400-e29b-41d4-a716-446655440000
  name: my-app

reviewers:
  primary: codex    # Default reviewer (required)
  secondary: gemini # For second opinions (optional)

How It Works

glee init
    ├── Creates .glee/config.yml
    └── Creates .mcp.json (MCP server registration)

claude (start in project)
    └── Reads .mcp.json
        └── Spawns `glee mcp` as MCP server
            └── Claude now has glee_* tools

Documentation

• docs/VISION.md - Project vision and design principles
• docs/PRD.md - Full product requirements
• docs/subagents.md - Subagent orchestration design
• docs/workflows.md - Agents & workflows design
• docs/arbitration.md - Review feedback system

Development

# Clone the repository
git clone https://github.com/GleeCodeAI/Glee
cd Glee

# Install dev dependencies
uv sync

# Run CLI during development
uv run glee --help