AI that watches your screen and tells you what to do.
Share any window. Ask anything. Tunjuk reads what you see and answers out loud — privately, on hardware-isolated GPUs, billed to your own Chutes wallet.
Bring your own compute
OAuth into Chutes once. Every inference call is billed to your wallet, not to anyone else. Tunjuk holds zero credits and never pays for your usage.
Confidential by hardware
When a TEE-attested model is available, we pin it. Your shared screen is processed inside an Intel TDX enclave that operators cannot read into.
Nothing to install
Web-native. Browser screen-share, voice-in, voice-out — no extensions, no native app, no permissions you do not grant in the browser itself.
How it works
Four steps from a shared screen to a spoken answer.
Sign in with Chutes
One PKCE OAuth handshake, scoped to chutes:invoke. Tunjuk never sees a password and holds zero credits — every call is billed to your own Chutes account.
Built on what you can verify.
Every answer comes with the model id Tunjuk picked, time-to-first-byte, total stream duration, and a verifiable Confidential Compute badge when the inference ran inside an Intel TDX enclave.
- ✓Picked-model label on every answer.
- ✓TTFB + total stream timer.
- ✓Confidential Compute attestation badge (proves environment, not correctness).
- ✓No data persisted: stateless functions, no DB, no logs.
- ✓MIT-licensed. Fork it and deploy your own in one click.