Privacy policy

Scope

This draft covers (1) this website and (2) the DOT404 desktop app for Windows. It describes data flows as implemented in the current build — not aspirations.

This website

• No analytics, no cookies, no tracking pixels. The pages are static files; nothing measures your visit.
• Fonts. The site self-hosts its typefaces (Inter and Orbitron, SIL Open Font License — license texts ship with the font files) on the same server as the pages. Your browser makes no requests to Google Fonts or any other font CDN.
• Hosting logs. Whoever hosts the site may keep standard server logs. Draft item: name the host and its retention.

The desktop app — stored on your device

• Conversations (questions and replies) in a local SQLite database in your OS app-data folder.
• Preferences: theme, sidebar state, Amethyst position and visibility, teaching mode — in local storage / the app’s settings file.
• Screenshots and microphone audio are never stored — screenshots exist in memory for one answer, audio until transcription.

The desktop app — what leaves your device

• AI requests you trigger: your message plus bounded recent context (up to 40 messages, 12,000 characters each, 100,000 total) sent to DOT404’s gateway, which forwards to the AI provider for that task.
• Screen questions: one screenshot per Send, sent with your question; for grounded questions the same bytes may also be sent to a second provider (currently Google Gemini, with Cloudflare Workers AI as fallback) to localize targets.
• Voice: the recording is sent for transcription and discarded; only the transcript and reply are saved. Speech requests send reply text for synthesis.
• Attachments: files are parsed on your device; extracted text (≤ 6,000 characters per file) and any images you pick are sent with your prompt.
• Account: sign-in data (email, or Google/Apple profile if you use those) is processed by Clerk as the authentication provider; your user id is used to scope your data.
• Backend roles: Supabase hosts the gateway functions and stores account-level metadata (usage/quota and subscription roles). Chat content is not written there in the current build.

How your content is used

• No training. Your prompts, screenshots and attachments are never used to train models.
• No ads, no profiles. Nothing here feeds advertising or cross-site tracking.
• No background sensing. Capture requires your explicit Send; recording requires your held button.

Retention

• Local conversations: until you delete them (deletion is immediate and durable) or uninstall/removes the app’s data.
• Screenshots / audio: not retained by DOT404 beyond the request.
• Provider-side: request transit may be subject to each provider’s own retention. Draft item: verify and link the retention terms of every live provider (Groq, OpenRouter, Google, Cloudflare) — zero-retention is NOT currently guaranteed.