An AI software-engineering agent that runs in a sandboxed cloud environment: a shell, an editor, a browser, git/GitHub, deployment previews, and evals — all visible and steerable from a web UI.
| Service | Job | Stack |
|---|---|---|
frontend/ |
The UI (session, terminal, editor, browser view, approvals) | React 19 + Tailwind 4 + Shadcn (bun) |
orchestrator/ |
The brain: agent loop, planning, memory, evals | Python 3.11 + FastAPI (uv) |
sandbox-gateway/ |
The body: PTY bridge, WebSocket fan-out, per-sandbox daemon | Rust + axum + tokio |
sandbox-fleet/ |
Sandbox lifecycle: containers/microVMs, warm pool, placement | Python + Docker/Firecracker SDKs |
infra/ |
Postgres, Redis, Mongo, MinIO (S3), egress proxy + docker-compose | Docker Compose |
cp infra/.env.example infra/.env # once
make init # boot databases + sandbox egress (docker compose up -d)
make dev-orchestrator # terminal 1 — API + agent brain (http://localhost:8000)
make dev-gateway # terminal 2 — Rust gateway
make dev-fleet # terminal 3 — fleet manager
make dev-frontend # terminal 4 — UImake test # all service tests
make ps # infra healthSee docs/ — the full theory curriculum (7 parts, 51 chapters) that drives this build:
foundations, sandbox isolation, tools, platform architecture, evals, security, build plan.