Browser AI Free
Runs on this device. Nothing leaves your computer.
- Gemini Nano built into Chrome and Edge
- Summaries, translation and Q&A
- Zero setup, zero cost, works offline
AI assistant
Understand documents in seconds. On your terms.
Shift To PDF runs AI where it makes sense for you: free and private on your own device by default, frontier cloud models when you want them. Every engine uses the same grounded prompts and cites pages.
Device-first by designNo file ever uploadedFree engines for everyone
Four engines, one assistant
Pick per task, or let Automatic choose the cheapest private engine that is ready on your device.
On-device model Free
Runs on this device via WebGPU. Works offline after a one-time download.
- Open models (Qwen3, Llama 3.2) on your GPU via WebGPU
- One download of 0.9–2.6 GB, cached by the browser
- Reads long documents in parts and merges the result
Cloud model Included in Pro
Extracted text is sent to our server and a model provider. Nothing is stored.
- Frontier-class models hosted by us
- Only extracted text is sent; nothing stored
- Included in Pro with a daily allowance
Your own key Billed by your provider
Text goes straight from your browser to the provider you configured.
- Gemini, Groq, OpenAI, OpenRouter, DeepSeek, Anthropic or self-hosted
- Key stays in your browser; calls go direct
- You pay your provider, we never see the key
Tools
Each one works with every engine.
Summarize PDF
Get an executive summary, key points and action items from any document.
Chat with PDF
Ask questions about your document and get answers grounded in its text.
Translate PDF
Translate document text into 30+ languages while keeping its structure.
Extract data
Pull tables, invoice fields and key facts into clean JSON or CSV.
How it protects your documents
Text is extracted locally
The PDF is parsed in your browser. The file itself never leaves your device, whichever engine you use.
On-device stays on device
Browser AI and the on-device model never make a network request with your text. Offline works.
Cloud asks first
Before any text goes off-device you see exactly what will be sent and approve it for the session.
Nothing is stored by the app
No prompt, text or answer is written to our servers. Save results locally if you want to keep them.
Scanned PDFs have no selectable text. Run OCR PDF first (on device), then use any AI tool on the result. Manage engines, models and keys in Settings → AI.