Unicorneatrowel/Gemini-Config-Checker-

What it does

Gemini Config Checker is a desktop, web-server and Android tool that evaluates proxy configs by asking a more realistic question than "does the route reach Google?": does Google's region wall let a signed-in user into AI Studio and Gemini?

The README argues that plain HTTP probes report healthy from every country, because Google decides availability per signed-in account after the page loads. The tool therefore opens AI Studio and Gemini in a dedicated browser profile, reads the result (an AI Studio redirect to /docs/available-regions, or the "Gemini isn't currently supported in your country" message), and reports each config as healthy, blocked or errored.

Two modes cover different jobs. Simple tests a pasted list, subscription URL, base64 blob or a ready-made free list and reports which configs work. Advanced chains free configs behind a base config that already works from your machine, so you can find a different exit that Google accepts. Both modes support copying share links, a base64 subscription, or a ready-to-import Xray config; the export can include the anti-DPI finalmask for PattN/PattNG or strip it for stock Xray/v2rayN.

Quick start

1. Get a build from the Releases page: there are Windows, macOS and Linux desktop packages, Android APKs and web-server binaries. Releases include SHA256SUMS.txt.
2. Install a Chromium-based browser (Chrome, Edge, Brave or Chromium) if you want the signed-in region check. Android only supports the connectivity-only mode.
3. Launch the app, open the Scan tab, paste configs one per line (vless://, vmess://, trojan://, ss://), a subscription URL, or base64 text. Leaving the box empty tests a ready-made free list.
4. Click Sign in with Google once in the private browser window that opens; the window closes on success.
5. Press Start scan and read the Results tab. A config is healthy only if it passes your "healthy" rule, which defaults to both AI Studio and Gemini.

For the server build, the command ./gemini-config-checker serves the UI at http://localhost:8787; flags include -addr, -no-open and -version. Building from source uses make web, make desktop or make android with Go, plus Wails and the platform webview for desktop builds (build instructions).

Constraints and caveats

• Region checking needs a graphical session and a supported Chromium browser; without one, switch the check type to connectivity only, which does not verify region availability. Browsers with a Flatpak install cannot be driven by the app.
• Each browser check needs roughly 200 MB of RAM; the default is 4 simultaneous checks and 60 configs per scan. Region checks take about 20–40 seconds per config.
• Verdicts are samples, not guarantees: a Cloudflare Worker exit can change between connections, so a config can pass and later fail. The README recommends re-testing any config you keep.
• The anti-DPI profile (fragment masks, cipher suites, ALPN, clean IP) is aimed at filtered networks and adds latency elsewhere. Stock Xray cores cannot parse the fork's per-segment fragment settings, so the export has a PattN and a standard variant.
• The web UI binds to localhost but has no authentication; expose it only through an SSH tunnel.
• Sign-in cookies are kept in memory in an isolated profile. Free configs are operated by unknown third parties, so avoid sensitive traffic through them.

The README labels the project a 0.1.0 pre-release exercised end-to-end only on Windows; macOS, Linux and Android builds came from CI and were not tested on real devices (platform notes). The listed license is MIT for the project, with Xray-core under MPL-2.0 and the bundled fonts under their own licenses.

HotRepo's take

The repository appeared on the HotRepo new-project list on 2026-11-15, recording 192 stars in its first observed growth window and no forks. That is an early single-window signal rather than a track record, and the lack of forks means there is no visible ecosystem of derivatives or contribution yet. The niche is narrow and practical: people who rely on proxy configs for AI Studio access and want a verdict closer to what a signed-in user sees. The main open question for readers is distribution rather than concept — the README's top download banner points to a Dropbox .rar separate from the GitHub releases, and the release artifacts are unsigned binaries, so checking SHA256SUMS.txt and preferring the GitHub release page is the sensible path.