Autonomous Multi-Agent Development Framework

Kollaborate orchestrates multiple AI agents working in parallel on complex software projects. It's a general-purpose LLM CLI management framework designed for massive parallelization and autonomous operations.

Remote Installation (Recommended)

curl -fsSL https://raw.githubusercontent.com/maarco/kollaborate/main/install.sh | zsh
source ~/.zshrc  # or ~/.bashrc

Local Installation

git clone https://github.com/maarco/kollaborate.git
cd kollaborate
./install.sh
source ~/.zshrc  # or ~/.bashrc

The same script works for both remote and local installation, automatically detecting the method and installing to ~/.kollaborate.

cd your-project
kollaborate init

kollaborate start 3 2  # 3 task agents, 2 spec agents

• Multi-LLM Support: Switch between Claude, Gemini, Crush, Codex, OpenCode, Cursor with one command
• 17 Typed Task System: Specialized prompts for features, refactors, bugs, tests, docs, performance, architecture, security, hotfixes, migrations, integrations, chores, experiments, UX, validation, workflows, and exploration
• Parallel Agent Execution: Multiple agents work simultaneously on different tasks
• Automatic Spec Generation: Agents create detailed specifications before implementation
• Activity Monitoring: 3-strike idle detection with automatic cleanup
• Task State Management: NEW → WORKING → DONE/QA workflow
• Global Installation: Runs from ~/.kollaborate without per-project file copying
• TMUX Orchestration: Full agent lifecycle management with tglm, tlist, tcapture, tmsg, tstop

# Initialize in current directory
kollaborate init

# Switch LLM backend
kollaborate use              # Show current backend and options
kollaborate use gemini       # Switch to Google Gemini
kollaborate use claude       # Switch to Anthropic Claude
kollaborate use crush        # Switch to Crush AI
kollaborate use codex        # Switch to Codex

# Start daemon with 3 task agents and 2 spec agents
kollaborate start 3 2

# Check status
kollaborate status

# Add tasks with types
kollaborate add feature "User authentication" src/auth.js
kollaborate add test "API integration tests" tests/api.test.js
kollaborate add bug "Fix memory leak" src/pool.js

# Show help
kollaborate help

1. NEW tasks trigger specification agent creation
2. Spec agents generate detailed requirements (minimum 50 lines)
3. Task agents spawn once valid specs exist
4. Activity monitoring tracks agent progress via output hashing
5. Cleanup occurs when tasks transition to DONE/QA

NEW → WORKING → DONE/QA
  ↓       ↓
BLOCKED  ← (recycled if agent goes idle)

~/.kollaborate/           # Global installation
├── kollaborate           # CLI entry point
├── kollaborate.sh        # Agent watcher daemon
└── kollaborate.md        # Framework documentation

your-project/
├── TASK_TRACKING.md      # Task state manifest
└── specs/                # Generated specifications
    ├── F01-user-auth.md
    ├── R02-api-refactor.md
    └── T03-integration-tests.md

Kollaborate supports multiple LLM CLI tools through the unified glm wrapper:

Switch backends anytime:

kollaborate use gemini    # Switch to Gemini
glm "your prompt"         # Uses current backend

Kollaborate uses TMUX-based agent orchestration with these commands:

• tglm "$agent_name" "$prompt" - Spawn agent session
• tlist - List active agents
• tcapture "$agent" - View agent output
• tmsg "$agent" "$message" - Send message to agent
• tstop "$agent" - Terminate agent

All commands work with any configured LLM backend.

Edit kollaborate.sh to customize:

MAX_AGENTS=3              # Maximum concurrent task agents
MAX_SPEC_AGENTS=2         # Maximum spec generation agents
CHECK_INTERVAL=60         # Monitoring cycle (seconds)
NEW_TASK_REQUIRED=5       # Minimum task queue size
SPEC_TIMEOUT=300          # Spec generation timeout (seconds)

• zsh or bash shell
• External agent system (e.g., tglm, Claude API, OpenAI API)
• Standard Unix utilities (sed, grep, awk)

MIT

Contributions welcome! Please submit issues and pull requests to the GitHub repository.