How Lab works

You send code. Lab runs it. You get a receipt.

That’s the whole idea. Your agent (or script, or curl command) sends JavaScript to Lab. Lab runs it inside a Cloudflare sandbox — a fresh, isolated environment that starts in milliseconds. When you run code on Lab, the receipt JSON is available at GET /results/:id.json. Humans can open GET /results/:id in the public app viewer. Successful runs include full step data. Failed or aborted runs include the error and reason; per-step detail may be partial.

Permissions start at zero

Your code can’t do anything dangerous by default. It runs in a locked-down sandbox with no network access, no storage, no AI calls — nothing.

You unlock what it needs by declaring permissions:

If you don’t declare it, the code can’t use it. There’s no “grant all” shortcut.

Multi-step pipelines

You can run multiple pieces of code in sequence. The output of each one becomes the input of the next:

Step 1: Fetch data     →  { users: [...] }
Step 2: Transform it   →  { cleaned: [...] }
Step 3: Validate it    →  { valid: true, count: 42 }

Each step runs in its own sandbox. Each step can have different permissions. Step 1 might need kvRead. Step 2 might need nothing. Step 3 might need d1Read.

Code can launch more code

If you grant the spawn permission, your code can launch additional sandboxes — useful for parallel work or breaking a problem into parts.

There’s a depth limit. Each level down gets the same or fewer permissions than its parent, never more. At depth 0, spawning is blocked entirely.

What gets saved

Every run saves a JSON result document. GET /results/:id.json is the canonical machine path, and GET /results/:id is the human-facing viewer on the public app.

For successful runs, it contains:

• The code that ran in each step
• Each step’s input and output
• Timing for each step

For failed or aborted runs, it contains the top-level error and reason. Per-step detail depends on where the run stopped and may be partial or empty — see Failures & step data.

See the full schema for details.

The four packages

Under the hood (for contributors)