AgentCraft Documentation
Orchestrate your AI agents with an RTS game interface.
Welcome to AgentCraft
AgentCraft transforms your AI coding workflow into a real-time strategy experience. Watch your agents come alive in an RTS game interface — track and summon agents and subagents, command them, and manage them. Recycle with the intuitive interface you know and love.
What is AgentCraft?
AgentCraft is an open-source tool that reimagines how you interact with AI coding agents. Instead of a plain terminal, you get a full RTS-style command center where each AI agent is a hero unit on a 3D map, complete with missions, fog of war, skill scrolls, and achievements.
AgentCraft works by installing lightweight hooks into your agent CLI (Claude Code, Codex, OpenCode, or Cursor). These hooks report session events to the AgentCraft server, which renders everything in a browser-based game UI.
New in 0.5.0
- Campaigns — Turn a feature spec into a reviewed plan, dependency-aware agent tasks, integrated verification evidence, and a final user-controlled merge
- Multi-repository workspaces — Start AgentCraft from a parent folder and choose among discovered Git repositories or worktrees when spawning a hero
- Loop-to-Campaign delegation — Let a recurring loop hand a concrete feature or fix to the Campaign planner without creating duplicate work
- Safer uninstall and restore — Preview managed removals, retain timestamped backup manifests, restore a chosen session, or explicitly purge user data
- Codex session isolation — Keep internal Codex threads resumable in AgentCraft without filling the Codex desktop app's normal task list
- Provider-aware handoff and fork — Use H and Shift+H with external heroes while preserving the source hero's executor
Key Features
- Multi-Agent Heroes — Each AI agent appears as a hero on the map with real-time status tracking
- Campaigns — Plan, review, dispatch, verify, and merge multi-agent feature work with persistent evidence
- Multi-Repository Workspaces — Discover nested Git repositories and choose a repository or worktree per hero
- Agent Teams — Display Claude Code team configurations with grouped parties
- Alliance Hall — Collaborative multiplayer: create a shared room and see remote agents from other developers on your map in real time
- Mission Tracking — Completed missions persist across restarts with AI-generated summaries
- Fog of War — Toggle with V to visualize which areas have active heroes
- Skill Scrolls — Collectible scrolls on the map that install real agent skills from skills.sh
- Achievements — Tiered unlock system with shareable trophy cards
- Channels — DM your heroes from Telegram or Discord to send prompts, approve plans, and grant permissions
- Remote Access & Mobile — Secure tunnel sharing with installable mobile PWA and push notifications
- Voice Input — Speech-to-text in the composer with auto-send on silence
- Isolated Containers — Run agents in Docker or Apple Containers with full network isolation
- Scheduled Tasks — Create recurring agent tasks with cron-like intervals
- Integrated Terminal — Full PTY terminal in the bottom HUD with multi-tab support
- File Explorer — Browse project files in the side panel with search and IDE integration
- Git Worktrees — Spawn heroes in different worktrees directly from the UI
- Git Management — Built-in stash and branch management
- Music — Ambient soundtrack toggle
- Race Skins — Choose between Orc, Human, Elf, and Undead factions with unique buildings, heroes, terrain, and portraits
Quick Start
npx @idosal/agentcraftThis runs the guided setup: detects your agent CLIs, installs hooks, starts the server on port 2468, and opens the browser UI.
Run it inside one Git repository, or from a parent directory containing several repositories. Spawn Options lists the discovered working directories.
Head to Getting Started for a full walkthrough.
Supported Agents
- Claude Code — Full integration with Anthropic's CLI agent (primary)
- Codex — OpenAI's coding CLI with reasoning effort control
- OpenCode — 75+ models from multiple providers
- Cursor — Cursor's agent mode via CLI
- OpenClaw — Experimental
Learn the 0.5.0 Workflow
- Run a Campaign
- Use scheduled loops and Campaign delegation
- Launch a multi-repository workspace
- Manage install, uninstall, restore, and logs
Community
-
Discord — Get help and share feedback
-
Twitter / X — Follow for updates