OpenMagicPointer v0.1.1 — first-launch setup window

Small but meaningful UX upgrade — you no longer need to set an environment variable before first launch.

What's new

• First-launch setup window. On a fresh install with no key, the app now opens a small settings panel where you pick the provider (Anthropic Claude or OpenAI-compatible), endpoint URL, model, and paste your key. No more "error dialog → quit."
• OS-encrypted secret storage. Keys entered through the setup window are encrypted via Electron's `safeStorage` (DPAPI on Windows, Keychain on macOS, libsecret on Linux) and kept in a separate `secret.bin`. `config.json` continues to never contain your key.
• Settings… in the tray menu. Reopen the setup window any time to switch backend, change model, or rotate the key. Changes apply live — no restart.
• Smarter local-server detection. When you point at `localhost`/`127.0.0.1`/`*.local`, the setup window tells you no key is needed (for Ollama / LM Studio / llama.cpp / vLLM).

Heads-up about SmartScreen

Releases are not yet code-signed, so Windows SmartScreen will warn on first run. Workaround:

1. Verify the file against `SHA256SUMS.txt` (attached).
2. Right-click the `.exe` → Properties → tick Unblock → OK, or click More info → Run anyway on the SmartScreen dialog.

Working on a proper signing pipeline (likely SignPath.io's free OSS plan) for the next release.

Downloads

Artifact	When to pick
`OpenMagicPointer-Setup-0.1.1-x64.exe`	Most Windows 10/11 PCs (Intel/AMD 64-bit).
`OpenMagicPointer-Setup-0.1.1-arm64.exe`	Windows on ARM (Surface Pro X, Copilot+ PCs).
`OpenMagicPointer-0.1.1-x64-portable.exe`	No-install, run-from-anywhere x64 build.
`OpenMagicPointer-0.1.1-arm64-portable.exe`	No-install ARM64 build.

Full changelog: https://github.com/mengzili/openmagicpointer/blob/v0.1.1/CHANGELOG.md

Full Changelog: v0.1.0...v0.1.1

OpenMagicPointer v0.1.0 — initial public release

First public release. Open-source, proactive on-screen hints for Windows — an MIT-licensed, bring-your-own-VLM alternative to closed assistants.

Highlights

• 🧠 Pluggable VLM backend. Anthropic Claude (native SDK, prompt-cached) or any OpenAI-compatible Chat Completions endpoint — OpenAI, Azure, OpenRouter, Groq, Gemini's OpenAI-compat endpoint, and local servers (Ollama, LM Studio, llama.cpp, vLLM). Local URLs skip the API-key check.
• 🟦 Tiny floating bubble that appears near your cursor, only when a hint would clearly help, then fades.
• ⏱️ Conservative throttle: idle threshold, typing detection, minimum interval, and a 32×32 perceptual fingerprint that skips API calls when the screen hasn't meaningfully changed.
• ⌨️ Global hotkeys: F8 ask now, F9 pause/resume, Ctrl+Shift+F12 quit.
• 📌 System tray with pause/resume, "hint now", quit.
• 🖼️ Multi-monitor aware overlay with edge-flipping bubble placement.
• 🧪 46 unit + integration tests covering pure throttle/positioning logic, both VLM backends, and config round-trips.

Downloads

Artifact	When to pick
OpenMagicPointer-Setup-0.1.0-x64.exe	Most Windows 10/11 PCs (Intel/AMD 64-bit).
OpenMagicPointer-Setup-0.1.0-arm64.exe	Windows on ARM (Surface Pro X, Copilot+ PCs).
OpenMagicPointer-0.1.0-x64-portable.exe	No-install, run-from-anywhere x64 build.
OpenMagicPointer-0.1.0-arm64-portable.exe	No-install ARM64 build.

Verify with SHA256SUMS.txt (also attached).

Notes

• Installers are not yet code-signed. SmartScreen may warn on first run — verify via SHA-256 before running.
• Provide an API key via ANTHROPIC_API_KEY / OPENAI_API_KEY env var, or %APPDATA%\OpenMagicPointer\config.json. See the README for per-backend examples (OpenAI, Ollama, OpenRouter, Gemini-OAI-compat, …).

Privacy

• Captures are sent only to the backend URL you configured. With a local backend (e.g. Ollama), nothing leaves your machine.
• Identical screens are deduped locally via perceptual fingerprint before any network call.
• See README → Privacy and SECURITY.md.

Full Changelog: https://github.com/mengzili/openmagicpointer/commits/v0.1.0