1 · Your text
2 · Pick a task
3 · Engine preference
Auto routes short, private tasks to Gemini Nano on your device and escalates long documents & heavy generation to cloud Gemini — via Firebase AI Logic InferenceModes.
🔒 In Auto mode, short tasks never leave your device. Your text and history are stored only in this browser (localStorage).
Results
🚀
Paste some text, pick a task, and hit Run.
Every answer shows which engine produced it — and why it was routed there.
Hybrid routing log
Saved sessions (this browser)
How the hybrid routing works
StudySprint is built on Firebase AI Logic with hybrid inference for the Web. One SDK call — model.generateContent() — can be served by two engines:
- On-device: Gemini Nano, running inside Chrome via the Prompt API. Instant, free, offline, and private — your text never leaves the laptop.
- Cloud: Gemini (via the Gemini Developer API backend) for long documents and heavy structured generation.
The router picks an InferenceMode per task: PREFER_ON_DEVICE for short tasks, PREFER_IN_CLOUD for long/heavy ones, and ONLY_ON_DEVICE / ONLY_IN_CLOUD when you force an engine. If Firebase isn't configured yet, the app runs in demo mode: on-device tasks call the same Prompt API directly, and cloud-bound tasks degrade to a clearly-labelled local extractive fallback — never a fake AI answer.
Requirements for on-device inference: Chrome 139+ on desktop with chrome://flags/#prompt-api-for-gemini-nano-multimodal-input enabled (the model downloads automatically on first use).