Catbee
Open-source tools for hyper-personal AI — every project is designed to be self-hosted, fully controllable, and yours to own. Run entirely offline, mix in frontier intelligence when you need it, and keep your data where it belongs.
Open-source tools for hyper-personal AI — every project is designed to be self-hosted, fully controllable, and yours to own. Run entirely offline, mix in frontier intelligence when you need it, and keep your data where it belongs.
Catbee is a collection of open-source projects at the intersection of local-first AI and infrastructure ownership. Every tool is designed to run on your hardware, with your data, on your terms.
Every project runs locally by default. No telemetry, no accounts, no cloud dependency. Your prompts, your metrics, your documents — they stay on your machine.
Projects share storage formats (MooFile), API conventions (OpenAI-compatible), and configuration patterns. Mix and match tools without integration friction.
Start fully offline with llama.cpp models. Connect frontier APIs when you want more capability. Every project supports the full spectrum — your choice, not the vendor's.
From AI agent harnesses to embedded databases — every Catbee project is a self-contained tool you can run today. Click any project name for its dedicated site or GitHub repo.
Pengy is an AI agent that lives on your machine — not in the cloud. It reads your files, runs your bash commands, searches the web, writes code, and orchestrates your workflows. With 15 built-in tools and an extensible skills system, Pengy turns any LLM into a practical, local-first assistant. Available in Python (reference), Rust (high-perf), and C++ (leanest) — all sharing the same chat history and settings.
fast or smart for automatic routing and fallback.
Track token usage and cost per model per day, with matplotlib-powered charts on the reports page.
Add, edit, and delete models through the web UI — no config files to hand-edit.
AI shouldn't mean giving up control of your data, your infrastructure, or your privacy.
Every Catbee project runs without phoning home. There are no analytics pings, no mandatory accounts, no "sign up to continue" gates. Your usage patterns, your prompts, your documents, and your metrics belong to you — period.
From the LLM backend to the storage layer to the user interface, every component is open-source and self-hostable. Run entirely offline with llama.cpp, connect frontier APIs when you need them, or build your own hybrid — the choice is always yours, never a vendor's pricing tier.
Each project does one thing well and shares common conventions — MooFile for storage, OpenAI-compatible APIs for LLM access, JSON for data exchange. This isn't a platform; it's a toolbox. Grab what you need, wire it up, and leave the rest.
Every Catbee project is actively used by its creator — Pat Wendorf, a Solutions Architect at MongoDB who builds AI tools for his own home network, his own workflows, and his own curiosity. If a tool isn't useful to the person building it, it doesn't ship.
Catbee projects are designed to be used independently — but they also compose into a coherent local-first AI stack.
The LLM Proxy routes requests to local or cloud models → Pengy orchestrates tools and skills → MooFile stores everything persistently → TheWatcher monitors the whole stack → all visible from a browser on any device.
Catbee is maintained by Pat Wendorf — a Solutions Architect at MongoDB who builds open-source AI infrastructure from his home in Midhurst, Ontario. Every project is MIT-licensed, dogfooded daily, and free to use however you see fit.