Pointr v0.3.0

Highlights

• Push-to-talk voice. Hold Ctrl + Win, say what you want, and let go. Your speech is turned into text on your own computer while you're still talking, so the question goes out the instant you release. Start with "agent" or "explain" to use those modes, or just address it: "Pointr, open Notepad and type hello". Filler words like "uh" and "hey" before the keyword are fine.
• Voice stays on your PC. Transcription runs on-device (NVIDIA Parakeet, a one-time ~670 MB download from Settings -> Voice). Audio is never uploaded or saved. Prefer no download? Choose My OpenAI key in Settings -> Voice instead.
• You choose how much it can do on its own. On-screen actions now has three settings: Off, Always ask (default), and Always allow, which skips the Enter prompt. Esc still stops any run.
• Faster questions. The screenshot now uploads in the background the moment you press the hotkey, so only a short reference goes with your question.

Fixes

• Drafted replies and any multi-line text now type out correctly. Previously a draft with a blank line in it came out as an empty line.
• Error messages now show over your real screen instead of turning the whole overlay white.
• Voice no longer misses the key release when you let go of Win before Ctrl.

Good to know

• First launch of voice needs the one-time speech model download (Settings -> Voice -> Download), unless you use your OpenAI key.
• Requires 64-bit Windows 10 (1903+) or Windows 11, plus the Microsoft Visual C++ Redistributable (most PCs already have it). If Pointr won't start and mentions MSVCP140.dll or VCRUNTIME140.dll, install the x64 redistributable.
• The installer is unsigned, so Windows SmartScreen may show an "unknown publisher" warning.
• Still bring-your-own-key: connect a Gemini or OpenAI key in Settings.

Setup guide: https://www.pointr.duckdns.org/setup.html

Pointr v0.2.0

Highlights

• OpenAI support, alongside Gemini — connect either key, or both, and switch anytime in Settings → AI model. Every feature works with both: answers, explain mode, agent tasks, multi-step tasks, and the GitHub/web-search/file tools.
• Pick any model, not just the default — the model list is fetched live from your own key (filtered to models that can actually read a screenshot), plus a "Custom model ID" box for anything not listed. Defaults are gemini-3.1-flash-lite and gpt-6-luna.
• Both providers are BYOK — no server-side fallback for either. Without a key connected, you get a clear "not configured" message instead of a request quietly running on someone else's account.
• Opt-in anonymous usage telemetry, off by default, asked once — see the new Privacy page linked from the app's setup guide.
• Backend retention hardening — shorter result expiry, log rotation, no Redis persistence.
• Security fix: rotated the shared client key after finding it had been hardcoded in a tracked source file; it's now read from a local, gitignored .env at build time.
• Multi-step reliability fixes from real test runs (Notepad save-flow, foreground timing).

Verification

• 32 Rust tests pass
• Live click-grounding test across both providers on synthetic screens (gpt-6-luna 11/12 targets, Gemini 10/12, ~4px median error for both)
• Verified live against the production backend: old client key rejected, new key accepted, both providers correctly refuse to run without a key

Pointr v0.1.1

Custom app/tray icon, redesigned Settings panel, startup splash screen, and a lightweight update-check-and-prompt. Requires your own free Gemini API key (add it in Settings after install).

Pointr v0.1.0

First public Windows installer. Requires your own free Gemini API key (add it in Settings after install) — see the landing page for details.