# FAQ

> @faustine-ai/browser-sandbox documentation · https://browser-sandbox.faustine.ai/docs/faq/ · source: README.md on GitHub (https://github.com/faustine-ai/browser-sandbox)

**My page has a strict CSP. What do I need?** `url` mode and one
`frame-src` entry for the origin that serves the document. Nothing else:
no `'unsafe-eval'`, no `worker-src`.

**Can I run it on Safari?** Yes; the suite runs in WebKit. Safari 16+.

**Several sandboxes?** As many as you like; each has its own frame and
queue. Runs on one sandbox are sequential.

**Should I reuse workers?** `workerPerRun: false` saves a few milliseconds
per run and keeps whatever the previous run left in the worker's global.
Use it only when runs trust each other.

**How do I read the logs?** `result.logs`, or live through `onLog`.
`toToolOutput` includes them as `logs` for the model.

**Can I debug the worker in devtools?** Yes: it appears as a dedicated
worker under the sandbox frame; the model's code is the body of the
compiled function.

**Why is there no fallback to `new Function` in the page?** Because a
fallback is the hole. If the sandbox cannot start, tell the model, do not
run the code.

**Does `pause` time out?** The run clock pauses while `onApproval` is
pending. Cut a stale wait with the run's `signal`.

**What does the model see when a limit hits?** `toToolOutput` words it:
`Denied (sandbox_limit): the code ran 10000 ms with no function call in flight`.