FAQ

No. It captures a single fresh screenshot only when you press Send on a screen question, uses it for that answer, and lets it go. Nothing is recorded in the background — there is no timer, watcher, or screenshot history in the app. Full rules: screen awareness.

Never. DOT404 points at things with arrows, boxes, circles and pointers drawn on a transparent overlay — your hands stay on the mouse and keyboard. There is no computer-use or automation capability, by design.

No. There is no wake word and no background listening. The microphone opens only while you hold the button (60-second cap), shows a visible recording indicator, and the audio is dropped right after transcription.

In a SQLite database on your device, in your OS app-data folder — not in an ad profile and not in DOT404’s cloud. What does leave the device is each AI request you send, through the secure gateway: privacy policy (draft).

No. Your prompts, screenshots and attachments are never used to train models, and chat history stays in the local database. Provider-side retention details for transit are listed as a draft item in the privacy policy.

Windows 10 and 11 (64-bit) today. macOS is on the roadmap; there is no Linux or mobile build.

The app is free to download and there is no paid tier announced. AI usage runs through the project’s gateway and provider tiers. Public pricing has not been set — treat any pricing you see elsewhere as unconfirmed until it appears on the download page.

To use AI, voice, or screen questions, yes — a free account (email, or Google/Apple sign-in). Signed out, the app still runs: the character, themes and local conversations work, and AI replies fall back to a clearly labeled local preview.

Yes — the first public build, v0.1.0, is on the download page: a free Windows 10/11 installer (~101 MB). It is not code-signed yet, so SmartScreen may warn on first run — choose “More info” → “Run anyway”.

Partly. The character, animations, themes and local history work offline; AI answers, voice and screen questions need a connection because they run through hosted gateways.

Yes — push-to-talk voice chat on Amethyst: hold the mic, speak, release. The transcript becomes your question and the reply is spoken back. There is also dictation into the main composer and read-aloud on any reply. Details: voice interaction.

That’s the point. Amethyst lives in a tiny always-on-top window you can drag anywhere, and its guidance is drawn in a click-through overlay above whatever you’re doing.

Press Ctrl+Shift+Space anywhere in Windows, or use the DOT Mini row in the sidebar. Your last position is remembered on this device, and hiding it at quit keeps it hidden next launch.

Honest answer: uncalibrated. Normal-sized targets land well; tiny or dense UI can be missed. When grounding fails, you still get the text answer and simply no visual — never a wrong arrow pretending to be right. Up to eight targets can be drawn at once, for eight seconds.

Text, vision, grounding and voice each route to a primary provider with a fallback — currently Groq, OpenRouter, Google Gemini and Cloudflare Workers AI. The exact list lives in one place and is published on AI models & BYOK.

Not yet — BYOK is planned, not implemented. When it ships, keys will go into your OS credential storage, never into the chat database or logs. There is no BYOK UI or provider list today.