A Windows overlay that sees what's on your screen, answers in place, and can act on your behalf — with your permission, every time.
-8b5cf6){kind=icon}
Website · Setup Guide · Privacy · Releases · Issues
Press Ctrl+Alt+Space anywhere on Windows. Pointr captures your screen, marks exactly where your cursor is, and answers whatever you type — in place, with a pointer to what it's talking about. Ask agent: <task> and it can act for you: type into a focused field, open an app, click through a multi-step task across windows and a real browser. By default it shows what it's about to do and waits for Enter; Settings → On-screen actions can switch that to "Always allow" (no prompt, Esc still stops it) or "Off".
- Direct Q&A — hotkey, ask, get an answer anchored to a point on screen. No region-select step required.
- Push-to-talk voice — hold
Ctrl+Win, speak, let go. Speech is transcribed on-device (NVIDIA Parakeet, one-time download) while you're still talking, so the question is sent the instant you release. Audio never leaves your machine. explain: <topic>— a narrated, drawn-on-screen walkthrough instead of a wall of text.agent: <task>— one confirmed action (type a reply, open an app) or a full autonomous multi-step run, with a live step trace and a History window to review past runs.- Browser automation — a local Playwright-driven browser for tasks that need one, clicking by accessibility role/name instead of guessed pixel coordinates.
- Tool-using agent — reads your GitHub repos/issues/PRs (read-only), searches the live web (Tavily), and reads local files including PDFs, all opt-in and configured per-user.
- Narration — offline, local Windows text-to-speech; no cloud call, no extra cost.
- Session memory — a short-lived, per-app conversation history so follow-up questions have context.
Pointr doesn't run on a shared, metered backend. Connect a Gemini key, an OpenAI key, or both, in Settings — every request runs on your own key, nothing shared. Pick any model either provider gives you: the model list is fetched live from your key (filtered to whatever can actually read a screenshot), plus a free-text box for any model ID at all. Switch providers or models anytime; every feature works identically on either one.
Download the latest Windows installer from Releases, run it, and connect an API key in Settings — see the setup guide for the two-minute walkthrough. The installer talks to Pointr's hosted backend by default; nothing else to run.
Requires 64-bit Windows 10 (1903 or later) or Windows 11, plus the Microsoft Visual C++ Redistributable, which most PCs already have. If Pointr won't start and mentions a missing MSVCP140.dll or VCRUNTIME140.dll, install the x64 redistributable and try again.
Pointr is two pieces: a Tauri (Rust) desktop client, and a FastAPI + Celery backend it talks to. You can run both locally against your own backend instead of the hosted one.
Desktop client (app/)
cd app
npm install
npm run tauri devBackend (backend/, via Docker Compose)
# root .env needs at minimum:
# GEMINI_API_KEY=... (or leave blank and add a key in Settings instead)
docker compose up -dBy default the client build targets http://localhost:8000 — set POINTR_ENV=prod in the root .env to build against a hosted backend instead (see app/src-tauri/build.rs).
Requirements: Rust (stable) + the Tauri prerequisites for Windows, Node.js (client build, and lazily for browser-automation tasks at runtime), Docker (backend).
The backend is a relay, not a store: screenshots, queries and answers pass through per request and aren't retained. See Privacy for exactly what's kept, for how long, and why.
Usage telemetry is opt-in, off by default, asked once. It's anonymous (a random install ID, no account) and never includes screenshots, queries, file contents, answers, or tokens — only which features get used and whether a request failed. Full details: privacy page.
Issues and PRs welcome. There's no formal contributing guide yet — open an issue for anything non-trivial before sending a large PR, so the approach can be agreed on first.