★ FEATURED
🐾 purr
A terminal chat app that turns your local Ollama install into a multi-agent Mac assistant. 5 built-in personas, 29 tools, dangerous-tool confirmation, streaming Markdown, multi-tab chats, @file attachments, yolo mode, and a hallucination detector that strikes through lies and locks the tab after 3 of them.
Default persona: Dross (◉_◉) on gemma4:12b. Try purr →
No cloud, no API keys, no telemetry. Everything runs on your machine.
NEW
⌨️ TerminalDeck
A single-window dashboard for all 157 CLI tools on your Mac. 78 installed (Claude Code, Codex, ollama, kubectl, ffmpeg, …) + 83 discoverable (fzf, bat, k9s, …). Click-to-install runs brew install live. Built-in pyte-backed tabbed terminal. Right-click any tool to launch in Terminal.app or Warp for long-running TUIs.
PySide6 + pyte. No Xcode. 97 MB arm64 .app at /Applications/TerminalDeck.app. See it in action →
Mac-native keyboard shortcuts: ⌘T/W/D/⇧R/1–9, ⌘F filter, ⌘, prefs. QSettings remembers everything.
raised-letters-clone
An attempt to rebuild a popular subscription-letter writing platform from scratch. Includes backend, frontend, and build scripts. Work in progress — the spec is written, the foundation is laid, and the design is rough around the edges.
vellum-rs
A terminal writing app for macOS and Linux. The killer feature: press Tab and your current line gets rewritten by a local Ollama AI model — grammar fixed, prose sharpened, voice preserved. No alt-tabbing, no copy-pasting, no losing your flow.
Default model: qwen3:4b — fast, cheap, genuinely good at prose. Configure any Ollama model you like.
OurNook
A private, beautiful AI companion that lives on your machine. Long-term memory that actually persists, voice conversations, and full control over what it knows about you. No cloud, no telemetry, no middlemen.
Built for the people who want an AI that knows them — without sending their life to a server farm.
LocalWrite
A native macOS desktop app for writing with AI. Paste text, pick a mode (Professional, Conversational, Scholarly, Creative, Concise), and a local LLM rewrites it in place. Five curated open-source models (Qwen 2.5 7B, Gemma 2 9B, Llama 3.1/3.2, Mistral 7B) — auto-downloaded on first use, then fully offline forever.
Bonus features: Document chat (RAG) — upload PDFs/DOCX/MD/TXT and ask questions grounded in your content. Summarize & Expand — one-click transformations. Quick refinements — "more formal", "shorter", etc. Export to TXT, Markdown, or DOCX.
PyQt6 + llama-cpp-python with Metal GPU acceleration. 1.0.1 fixes the PyInstaller spec to bundle llama's native dylibs.