Overview

SwarmForge is a sophisticated yet lightweight platform designed for orchestrating multiple AI agents, transforming them into a cohesive, professional software engineering team. It leverages tmux for session management and provides a structured environment for agents to collaborate on projects. SwarmForge focuses on disciplined workflows, ensuring agents communicate effectively through a robust handoff protocol and operate within defined roles and constitutions.

Key Features

• Disciplined Agent Orchestration: Manages communication and workflow between AI agents across different git worktrees.
• Config-Driven Swarm Topology: Define your agent team's structure and roles using a project-local configuration file (swarmforge.conf).
• Project-Local Roles & Constitution: Roles and their specific prompts are defined per project, allowing for tailored agent behavior. A layered constitution system provides a flexible way to manage project-wide and role-specific directives.
• Backend Flexibility: Supports various AI agent backends per role, including codex, claude, copilot, and grok.
• Observable & Lightweight: Runs locally within tmux and terminal windows, offering real-time visibility into agent operations without significant overhead.
• Robust Handoff Protocol: Employs a daemon-driven handoff system (handoffd.bb) for reliable inter-agent communication via validated files, replacing direct messaging.
• Git Worktree Integration: Automatically manages dedicated git worktrees for agents, ensuring isolation and organized project states.
• Terminal Automation: Integrates with common terminal emulators (macOS Terminal, Windows Terminal) for seamless window management.

Typical Use Cases

SwarmForge offers distinct workflow configurations to suit various project needs:

• two-pack (Quick Backend Workflow): Ideal for rapid implementation and refinement of backend tasks. This workflow focuses on fast coding with TDD and unit tests, followed by cleanup and hardening, omitting formal specification and QA for speed.
• four-pack (Compact Specification Workflow): Suitable for moderate projects requiring Gherkin specifications and architectural oversight without excessive agent specialization. It balances specification, coding, refactoring, and architectural review.
• six-pack (Full Workflow): Designed for major projects demanding comprehensive specification, upfront QA, backend verification, and detailed architectural consideration. Each major quality gate is managed by a dedicated agent role, ensuring maximum discipline and review coverage.