Gink

Think. Talk. Create. -- an AI assistant for macOS, Windows, Linux, Android and iOS.

Built with Python and BeeWare (Toga + Briefcase).

What Gink does

• Chat with streaming answers, chat history that saves itself, smart titles, drafts
• Voice: natural voices, voice conversations, and "Hey Gink" (wake word detected on the device)
• Does things: alarms, reminders, tasks, files (PDF, Word, Excel...), images, weather, currency and maths; on desktop also apps, Finder, Terminal, music, volume, screenshots, notes
• Morning briefing that picks up where you left off
• Sync across devices, Gink Pro with local payment methods, light and dark themes

Run it on your computer

python -m venv .venv && source .venv/bin/activate      # Windows: .venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env    # then add your keys -- .env is never committed
python -m ginkassistant

Build the app

Run Briefcase from this folder. Apps are built in build/ginkassistant/<platform>/....

python -m briefcase create macOS && python -m briefcase build macOS && python -m briefcase run macOS

• macOS 10.15 Catalina: use Briefcase 0.3.15 (newer Briefcase makes apps that need macOS 11+); requirements.txt picks the right version.
• Android and iOS are built on GitHub (Actions tab) -- see deploy/IOS_AND_WEB.md.
• Check your setup any time: python tools/check_project.py (--fix repairs it).
• New icon: replace icons/source-icon.png, run python tools/make_icons.py.

Tests

pip install -r requirements-test.txt
python -m pytest tests

Gink on the web

The Gink server also serves a web version at /app -- on your computer, http://localhost:8000/app while Gink is running. It's the app's own chat page plus a browser bridge (webapp/), built with python tools/build_web_app.py: sign in, chat, images, natural voices, dictation, and chats that sync with the desktop app. Alarms, tasks and computer control stay in the app.

Developer API

OpenAI-compatible, so the official OpenAI libraries work: base_url="https://<server>/v1", api_key="gink_live_...". Keys are created on /developers (shown once, stored hashed). POST /v1/chat/completions (streaming too), GET /v1/models, POST /v1/images/generations. Calls share the account's daily allowance; each key also has a per-minute limit (GINK_API_PER_MINUTE, default 30).

Owner dashboard

https://<server>/admin, opened with GINK_ADMIN_KEY (Render generates it): users, Pro, usage (messages, Claude, web searches, images, voice), a 14-day chart, payments, feedback, service status, and cost estimates from the prices you enter. python tools/grant_pro.py gives an account Pro. 10 wrong keys from one address lock it out for 10 minutes.

Website

The site lives in website/ (one shared layout in tools/build_website.py). Preview it with python tools/build_website.py and open website/_site/index.html. It's published on GitHub Pages automatically. To show Gink Pro prices, set them in website/assets/pricing.js.

Server

Gink's AI, voices, images, sync and payments go through your own server, so no keys are inside the app. Deploying it (Render, Docker) and its settings: deploy/README.md.

License

MIT -- see LICENSE.