Every project runs on its own private virtual machine — a real, secure computer in the cloud, not a shared browser tab. Install software, run real databases, start your app, and watch it come to life in a live preview.
Not a sandboxed tab and not a shared container — a dedicated machine that behaves like the one under your desk, minus the setup.
A space of its own
Your project, walled off from everything else.
Each project gets its own virtual machine with its own filesystem and network boundary. A runaway script or a risky command in one project can’t reach another, and nobody else shares your space. It’s yours until you delete it.
One isolated VM per project
Private filesystem and network
Your code is never used to train models
A real computer, not a toy
Install software. Run a database. Start your app.
This is a full Linux machine, so the agent can do what a developer does: install packages, run a real Postgres or Redis, spin up background jobs, and boot your dev server on 0.0.0.0 so the preview can reach it. No “that command isn’t supported here” surprises.
Real package installs and build tools
Run databases and background services
Start long-running servers and watch the logs
See it live
Your app, running, right next to the code.
A live preview sits beside the workspace and updates as the app changes — so you watch it come together instead of guessing. Ask the agent to try a flow and it can drive the preview itself — clicking, filling in forms, navigating between pages — then report back console errors and layout glitches it caught along the way, narrating each step live as it goes.
Live preview alongside your code
The agent can click through and test your app itself
Point at the preview to leave a note on the UI
Pick up where you left off
Close the tab. Come back. Nothing’s gone.
Workspaces pause when you step away and resume in seconds — exactly as you left them. Your files, your running services, and your chat history are saved with your account, so you can continue across sessions and devices without re-explaining the project.
Reopen a warm project in well under a second
Files and chat history saved to your account
Same state, on any device you sign in from
Experiment safely
Try the bold idea. Rewind if it doesn’t land.
Projects save restore points as you work. If an experiment goes sideways — a refactor that breaks the build, a dependency that fights back — roll the workspace to an earlier checkpoint without losing the rest of your progress. The freedom to try things is the whole point.
Automatic checkpoints as you build
Rewind the workspace to a known-good state
Guardrails flag risky moves before they run
What’s inside a workspace
Everything the build needs, in one place.
Open a project and it’s all there — the editor, the conversation, the running machine, and the controls to ship.
Files & editor
The full project tree on a real filesystem. Browse, read, and edit any file — the agent writes here and you can too.
Chat + plan mode
Describe what you want in plain English. The agent drafts a plan you can review before it touches a single file.
Logs & terminal output
Watch real commands run — installs, builds, test runs, server logs — streamed straight from the machine, not faked.
Secrets (encrypted)
API keys and tokens are encrypted at rest. The agent can use them through controlled tooling, but raw values never leak back into chat.
Skills & design packs
Reusable skills and design packs come along for the ride, so the agent builds the way you want from the first prompt.
Deploy & GitHub
Push to GitHub and ship to Vercel when you're ready. Both are explicit actions you trigger — never automatic.
Why a whole machine?
Most AI tools run your code in a shared, stripped-down sandbox — fine for a snippet, frustrating for a real app. Gate 15 gives every project a dedicated virtual machine instead, so the agent can build the way a developer would: install what it needs, run the real stack, and prove the result in a live preview.
That isolation is also what keeps your work safe. Because each project is walled off with its own filesystem and network boundary, nothing you build can reach another project, and your code is never used to train models. You decide when it leaves the workspace — read more about that on our security page, and about choosing the AI that does the building — Claude, GLM, GPT, or Gemini — on AI models.
AI app builder
Ready to build?
Start with $3 in trial usage. BYOK from $8/month. Included model credits from $20/month.