Fusion is a multi-node agent orchestrator designed to automate the journey from rough ideas to production-ready code. It targets developers and power users seeking to accelerate software development by leveraging AI for planning, execution, and review across distributed environments. The core benefit is automated, controlled, and efficient code generation and task management, managed from a single interface accessible from any device.

How It Works

Fusion orchestrates AI agents through a structured workflow: a planning agent first translates natural language task descriptions into detailed PROMPT.md files outlining steps, file scope, and acceptance criteria. Each task then undergoes a "Plan → Review → Execute → Review" cycle, with optional human approval gates at various stages. Tasks are isolated within individual Git worktrees (e.g., fusion/{task-id} branches) to enable parallel execution without conflicts. The system supports a multi-node mesh architecture, synchronizing state across laptops, servers, and mobile devices, and is model-agnostic, allowing integration with local or cloud-based AI models like Anthropic, OpenAI, and Ollama.

Quick Start & Requirements

• Primary install/run: npx runfusion.ai launches the dashboard. A one-line installer is available for macOS/Linux: curl -fsSL https://runfusion.ai/install.sh | sh. The CLI can be installed globally via npm install -g @runfusion/fusion.
• Prerequisites: AI providers (e.g., OpenAI, Anthropic, Ollama) need to be configured via settings or environment variables.
• Links: runfusion.ai →, Docs, Discord.
• Setup: An onboarding wizard guides users through AI setup, optional GitHub integration, and creating their first task.

Highlighted Details

• AI Planning & Workflow Gates: Tasks are planned by AI agents and executed through a rigorous Plan → Review → Execute → Review cycle, ensuring quality and allowing intervention.
• Worktree Isolation: Each task operates in its own Git branch and worktree, guaranteeing parallel execution with zero file conflicts.
• Multi-node Mesh: Seamlessly synchronizes task state, agents, logs, and diffs across desktop, mobile, and web interfaces.
• Agent Companies: Supports importing and running pre-built teams of over 440 agents across 16 companies for autonomous, long-running operations.
• Model Agnostic: Integrates with any pi-compatible model, including local Ollama instances and cloud APIs from OpenAI and Anthropic.
• Missions: Enables hierarchical planning (Mission → Milestone → Slice → Feature → Task) with autopilot and validation contracts.

Maintenance & Community

The project ships updates weekly and maintains an active Discord community. The project status is noted as "early preview."

Licensing & Compatibility

The project is licensed under the MIT license, permitting open-source use, self-hosting, and commercial integration without vendor lock-in.

Limitations & Caveats

The project is in "early preview" status. Several runtime plugins, including Hermes, OpenClaw, and Paperclip, are marked as "experimental," indicating that their APIs and wire formats may change in future releases.