A spatial, visual interface for managing multiple AI coding assistants.
AgentForge is a visual skin for the terminal experience. When you're running multiple AI agents (Claude, Codex, Gemini) across different projects, terminal tabs become chaos. AgentForge gives you a spatial way to organize them—like an RTS game, but for your AI workforce.
Each agent is a real terminal session. Not a simulation. When you type to an agent, you're typing to a real CLI. When you see output, it's streaming from a real process.
Before: 6 terminal tabs, no idea which is which
After: A map where you can see all your agents at once
# Install dependencies
npm install
# Start backend server
npm run server
# Start frontend (separate terminal)
npm run dev
# Or run both at once
npm startOpen http://localhost:3000, click "Enter The Forge", press N to summon an agent.
| Class | CLI | Model | Role |
|---|---|---|---|
| 🏗️ Architect | claude | Opus 4.5 | Strategic planning |
| 🧙 Mage | claude | Sonnet | Implementation |
| 🛡️ Guardian | codex | Codex | Code review |
| 🎨 Artisan | gemini | Gemini Pro | Design |
| 🔍 Scout | claude | Haiku | Research |
| ⚙️ Engineer | claude | Sonnet | Focused builds |
| Key | Action |
|---|---|
N |
Summon new agent |
| Click | Select agent |
| Drag | Selection box |
Option/Alt + Drag |
Rotate camera |
| Scroll | Zoom |
- VISION.md — What we're building and why
- ARCHITECTURE.md — How the systems work
- TERMINAL.md — The terminal experience
- CONTRIBUTING.md — For developers/agents
- Frontend: React, React Three Fiber, Zustand, Tailwind
- Backend: Node.js, WebSocket, node-pty
- Build: Vite, TypeScript
Imagine playing an RTS game, but instead of soldiers, you're commanding AI coding assistants. Position them spatially. See what they're doing. Talk to them like characters in a game—but get real terminal output.
This is a personal productivity tool for power users who run multiple concurrent AI sessions.
Built with Claude Opus 4.5