GitHub Sentinel is a self-hosted agent that monitors one or more GitHub repositories around the clock, detects newly opened issues, analyzes each one with a language model, and displays the results in a local web dashboard. It is designed to run continuously on a small machine such as a mini PC or home server running Windows or macOS. On a configurable schedule (defaulting to every 30 minutes), the agent polls the GitHub REST API for new issues and stores each one in a local SQLite database to avoid processing the same issue twice. For each new issue, it sends the text to a language model of your choice and gets back a summary, a category (bug, feature request, documentation, or question), a risk level (low, medium, or high), the source files most likely involved, and a proposed solution. The dashboard is a minimal React interface styled with terminal-inspired fonts. It supports filtering and searching through analyzed issues. The backend exposes a small API that lets you add or remove repositories to watch, trigger an immediate scan, and force re-analysis of specific issues. The project is built on Bun, a JavaScript runtime that also bundles the frontend and provides a built-in SQLite driver, so there are no external runtime dependencies beyond Bun itself. The language model endpoint is configurable to any OpenAI-compatible API, including local model servers on the same network, so all data can stay off the public internet. Service installation scripts are included for both Windows (using NSSM) and macOS (using launchd), with automatic restart on failure and log rotation. The full README is longer than what was shown.
← midudev on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.