A personal computer in the cloud — powered by an AI agent that actually uses it.
Join the waitlist at construct.computer.
Each user gets their own isolated virtual desktop: a real browser, a terminal, a file system, a code editor. The AI doesn't just answer questions about your computer. It operates your computer — navigating the web, running commands, managing files, writing code, joining meetings, sending emails — all while you watch in real time through your browser.
construct.computer is built on a simple premise: the most natural interface for an AI agent isn't a chat box — it's a computer.
Rather than translating user intent into API calls or tool schemas, the agent is given the same interface a human would use. It sees the screen, clicks, types, reads, and reasons. You can take over at any time, collaborate mid-task, or just let it run.
Every user's environment is fully isolated, persistent, and theirs. Files stay between sessions. The agent remembers things. Integrations (Google, Slack, email) are connected once and just work.
- Frontend — React + TypeScript + Tailwind, with a full desktop OS aesthetic (boot screen, lock screen, windowed apps). All web functionality powered by TinyFish
- Backend — Bun + Elysia, handling auth, container lifecycle, and WebSocket streams (screencast, terminal, agent events)
- Construct Agent — the AI agent runtime compiled to a binary, running inside each container. Connects to an LLM via OpenRouter, uses tools, and exposes an HTTP interface for the backend to communicate with
- Container — a Docker sandbox per user. Runs Chromium, a PTY shell, and a Desktop MCP layer that gives the agent fine-grained control over the browser
# Install dependencies and set up the environment
./scripts/setup.sh
# Start everything in development mode
./scripts/dev.shRequires Docker, Bun, and Node.js. See individual component directories for more details.
See LICENSE.