---

✨ Features

• 🔗 Connect GitHub with a personal access token — validated and stored in the macOS Keychain, never on disk.
• 📋 Multi-repo kanban — drag issues across Backlog → Todo → In Progress → In Review → Done. Statuses live locally, so your workflow isn't limited to GitHub's open/closed.
• 🪄 Create / update / close issues — writes straight through to GitHub.
• 🤖 AI done-check (BYOK) — an agent inspects merged PRs, commits, and comments, then reports done / partially done / not done / insufficient info — with evidence, confidence, and a suggested status. Works with any OpenAI-compatible provider.
• 📊 Dashboard & Analytics — live stats, a status donut, and per-repository / per-assignee breakdowns.
• 📬 Mailbox — issues assigned to you (or that you've claimed) in one place.
• 📝 Report — a one-click markdown summary you can copy out.
• 🎨 Gorgeous, keyboard-friendly UI — shadcn/ui with light/dark themes and drag-and-drop.

🖥️ Why tinyjs?

• ~6 MB shipped — no Electron, no bundled Chromium.
• No HTTP server, no ports — the UI and backend talk over a private Unix socket.
• Native macOS — Keychain secrets, real .app bundles, notifications.
• Local-first — your issues live in SQLite on your machine; your tokens live in the Keychain.

🚀 Quick start

# 1. install tinyjs
curl -fsSL https://tinyjs.app/install | sh

# 2. clone & run
git clone https://github.com/scarranca/issue-kanban.git
cd issue-kanban
npm install
tinyjs dev        # native window with hot reload
tinyjs build      # package dist/"Issue Kanban.app"

Then in the app:

Settings → Account (name + GitHub token) → Repositories (add repos) → AI (optional LLM key) → Sync.

🔑 GitHub token scopes

Create a token at GitHub → Settings → Developer settings → Personal access tokens.

Write access on Issues is only required for create / edit / close.

🤖 AI providers

Any OpenAI-compatible endpoint works. Defaults:

Base URL and model are editable per provider, and you choose the default used for done-checks.

🏗️ How it works

backend/          txiki.js backend — full system access, no ports
  main.ts         API surface (tiny.api.call)
  db.ts           SQLite (tjs:sqlite) — repos, issues, analyses
  github.ts       GitHub REST client (pagination + create/update/close)
  ai.ts           minimal OpenAI-compatible client (no SDK)
  agent.ts        done-check agent (prompt + verdict parsing)
src/              React 19 + shadcn/ui + Tailwind v4
  components/     sidebar, topbar, board, dashboard, analytics, mailbox, …

• The page and backend talk over a private Unix socket — no HTTP server, no ports.
• GitHub token + LLM keys live in the Keychain; board state in SQLite.
• The AI verdict never trusts a "closed" state on its own — it reads PRs, commits, and comments to decide if the issue is actually done.

🧭 Roadmap

• Click-through from the Analytics donut to a filtered board
• Export the report to a .md file
• Label-based status sync back to GitHub
• Markdown preview toggle while editing

📜 License

MIT © 2026 Issue Kanban contributors