An open-source control panel for coding agents. Run multiple AI CLIs in isolated, collaborative environments. Local. Lightweight. Yours.
Stackbox AI lets you run Claude Code, Codex, Gemini CLI, and other coding agents simultaneously — in isolated, collaborative environments called RunBoxes. Each RunBox has its own memory, its own git worktree, and its own container. All running locally on your machine.
No cloud. No accounts. No telemetry. Just your agents, your machine, your data.
Run Claude Code, Codex, Gemini CLI, and other agents simultaneously in split panels. Supports left, right, horizontal, and vertical layouts. Each panel is an independent terminal running its own CLI agent.
Named environments where multiple agent panels share the same memory and context. A RunBox called auth-feature can run a planning agent and an implementation agent side by side — both sharing the same context, aware of each other's progress.
Every RunBox automatically creates and manages its own Git worktree. Work on multiple branches simultaneously without stashing or context switching. When the RunBox is done — review, merge, close.
Each RunBox runs inside its own isolated Docker container. Agents can install dependencies, run servers, and modify files freely. Nothing touches your host machine. When the RunBox is closed, the container is destroyed and your system stays clean.
Inline browser panel for viewing running apps, testing UI, and checking localhost ports — all without leaving Stackbox.
Real-time diff view of every file created, modified, or deleted by agents. Line-level changes, per session, per RunBox.
Click any file in the File Changes Viewer to instantly open it inline — syntax highlighted, with diff highlighting showing exactly what the agent added or removed. No switching to VS Code. No losing context. Read-only and always in reach.
A persistent, shared memory system powered by LanceDB — embedded directly inside Stackbox, zero setup required.
Every RunBox has its own memory store. When an agent discovers something mid-session — a pattern, a bug root cause, an architecture decision — the Memory Layer captures it and makes it available to all other agents inside the same RunBox in real time.
- Capture — automatically extracts key discoveries from agent outputs
- Share — all panels inside a RunBox read from and write to the same memory
- Persist — memory survives when the RunBox closes, agents pick up where they left off
- Inspect — a dedicated Memory Panel lets you view, edit, pin, or delete any memory entry
Memory is stored as vectors using LanceDB, enabling semantic search — agents retrieve contextually relevant memories even when the exact wording differs.
| Layer | Technology |
|---|---|
| Frontend | Vite + React + TypeScript |
| Desktop wrapper | Tauri 2.0 |
| Backend | Rust |
| Terminal / PTY | portable-pty crate |
| File watching | notify crate |
| Memory / Vector storage | LanceDB (embedded, serverless) |
| Container isolation | Docker |
- Local first — no cloud, no accounts, no telemetry
- Lightweight — fast to install, fast to open, low memory
- Offline capable — works fully without internet
- Open source forever
- Privacy first — your agents, your machine, your data
- Your API keys — all integrations use your own keys, nothing proxied
# Clone the repo
git clone https://github.com/stackboxai/stackbox.git
cd stackbox
# Install dependencies
npm install
# Run in development
npm run devRequires Docker to be running for containerized RunBoxes.
Contributions are welcome! Open an issue or submit a PR on GitHub.
One RunBox. One context. Multiple agents. All in sync. Nothing touches your machine.