Paste untrusted or AI-generated code and see what it actually does in a disposable sandbox — before it touches your machine.
ℹ️ What is this? How it works & limits
SafeRun runs untrusted or AI-generated code in a throwaway cloud sandbox and tells you, in plain English, what it actually did — before it ever touches your machine. You can scan pasted Python, an npm package, or a PyPI package.
What is Solari? Solari provides the disposable, hardware-isolated cloud VM each scan runs in. Your code executes there, never on your computer, and the VM is destroyed right after.
Which AI writes the verdict? Claude, via the Anthropic API. You bring your own Anthropic key (default model claude-haiku-4-5 — the key is yours, so you control the cost). The behavior report and risk level are computed deterministically; Claude turns them into the plain-English verdict.
Limits & caveats:
- It runs your code, so very heavy or long-running code may hit a ~60-second timeout.
- Max input size is 200 KB (a few thousand lines). It's about run time, not line count.
- Python wheels run no install-time code, so they correctly show as clean.
- Shell-only install scripts (e.g.
curl) aren't traced at the syscall level; Node- and Python-level behavior is. - Your keys stay in your browser and are used only to run your scan — never stored or logged.
- Best-effort behavioral check, not a security guarantee.
SafeRun runs on your keys — stored only in this browser, sent only to run your scan, never saved or logged. You'll need a free Solari key and an Anthropic key.
- Get a free Solari keyopen getsolari.com ↗Sign up (the free tier includes starter credits), open the dashboard, and copy your key — it starts with
slr_live_. Paste it here:💡 Want stealth + more? On the Starter plan you can redeemSTARTER1MO-MKY4BNDKfor 1 month free — a public promo shared by Solari for the challenge. - Add your Anthropic key for the AI verdictopen console.anthropic.com ↗Claude uses it to write the plain-English verdict (default model
claude-haiku-4-5— the key is yours, so you control the cost). Starts withsk-ant-:
⚡ Powered by Solari — why it's great & what else you can build
SafeRun is one small example of what Solari makes possible. It gives AI agents real computers — browsers, sandboxes, and desktops — behind a single API key.
Why the platform is great:
- One API key for cloud browsers, code sandboxes, and full Linux desktops, all on one balance.
- Hardware-isolated micro-VMs that boot from a snapshot in about a second — nothing you run can touch your machine.
- Stealth cloud browsers that get past bot-detection, with managed proxies and CAPTCHA solving.
- A safe place to run untrusted or AI-generated code — exactly what SafeRun relies on.
Other things you could build on it:
- AI code interpreters that run the model's code safely.
- Web scraping & automation on sites that block ordinary bots.
- Computer-use agents that click through real GUIs.
- Ephemeral CI/test sandboxes and browser testing.
Explore the Solari cookbook and docs.