U
User
Free
Local Key Storage
Provider credentials are stored in this browser's local storage. Any XSS vulnerability could expose them. Use session-only mode when you do not want API keys, ChatGPT Codex tokens, or 𝕏/xAI tokens to persist.
Hover for details
On-device model (in the browser)
The tif1 model (a 1.7B fine-tune for F1 questions) runs on your GPU
through WebGPU — no key, no account. The model runs on your GPU: your questions and data go to no AI provider. Pick In-browser (WebGPU) in the chat's provider
list. Needs a recent Chrome or Edge on a desktop or laptop, about 1.4 GB of
GPU memory and a one-time download of about 1.14 GB (resumable; stored in this
browser). Your questions are still written to the site's anonymous question
log, as for every provider.
Asks the browser for its GPU and reads the model's 3 KB file list.
Custom providers
Bring your own endpoint — a local model (Ollama, LM Studio, vLLM,
llama.cpp) or a hosted OpenAI-compatible API. Stored in this browser (IndexedDB, encrypted with your device key); the key never leaves your device. For thinking models,
toggle think on the model chip.
No custom providers yet. Add a local model or a hosted OpenAI-compatible
endpoint.
Provider
Choose which AI service to use for generating SQL and answers.
Get a free Google AI Studio key
- 1 Open aistudio.google.com/apikey and sign in with Google.
- 2 Click “Create API key”, then copy the key.
- 3 Paste it below under “Add New API Key” and save.
Open Google AI Studio
Free tier available — no credit card required for most accounts.
AI Model
Pick the model that writes the SQL. With Answer model set to Fast, a smaller sibling writes the answer; Quality uses this model for both.
Add API key to see available models
Answer model
Your selected model always writes the SQL. Fast hands the written answer to a smaller sibling of it; Quality uses your selected model for the answer too.