Skip to content

v0.1.0 — first public release

Latest

Choose a tag to compare

@JeronimoRepetto JeronimoRepetto released this 24 Sep 22:01
b7aa085

First public release of Issue Classifier, live at https://issueclassifier.com.

Paste a GitHub repository and the app fetches its open issues and README, then rates every issue by criticality, complexity, effort and relevance, with a confidence for each answer. Filter, dismiss, sort by weighted priority and export the result.

Highlights

  • Three ways to rate issues
    • Jev (TypeSafe AI's decision model) with your own API key, one request per batch of issues.
    • A local model server on your machine: Kev / JevK5, or Laya (per issue, 512-token states), with OS-aware setup commands and one-click launchers for Kev.
    • An in-browser model on WebGPU through Transformers.js (a small placeholder model for now; WebAssembly fallback).
  • Triage view: search and multi-select filters (collapsible), relevance and confidence ranges, weighted priority sort, dismiss.
  • Export as plain text, Markdown or self-contained HTML.
  • Local-first privacy: keys live in memory by default (explicit opt-in to persist), analyses are stored in your browser, strict CSP, no third-party scripts.
  • Hosted on Cloudflare Pages with a hardened same-origin /jev proxy (path allowlist, origin check, rate limit, no logging, fails closed).
  • Link previews for chats and social networks, robots.txt, sitemap and structured data.

Quality

  • 2,100+ tests (Vitest), typecheck, hygiene checks (including pnpm audit) and build checks on every pull request.

Known limitations

  • The in-browser model is a generic placeholder, not a trained classifier yet.
  • Laya reads only ~512 tokens per issue, so README context is trimmed hard; compare it with Jev before relying on it.

MIT licensed. Issues and pull requests are welcome — see CONTRIBUTING.md.