Pawsitter
A private, local-first AI assistant that helps pet sitters follow the owner's instructions: meds, feeding, schedules and where things are.
About the project
Pawsitter is a private, local-first AI assistant that helps pet sitters follow the owner’s instructions: medications, feeding, schedules and where everything is.
The sitter opens the dashboard on their phone (on the home Wi-Fi) and asks things like “How much Ondansetron does Sonic get?” or “What do I need to do next?”. Answers come from a local LLM using only the owner’s own notes, with photos of the right pill bottle or pen. Nothing leaves the home network.
Features
- Ask anything: chat with a local LLM that answers only from the household’s information and says “I don’t know, contact the owner” instead of guessing.
- Knows the time: the current time, overdue and upcoming tasks and recent activity are in every prompt, so “what’s next?” just works.
- Photos in answers: knowledge items can have photos (the real medication pen, the cat food can) shown alongside matching answers.
- Daily checklist: tap “Done” on each task; the owner sees the activity history.
- Owner view: manage pets (with photos), schedules and household knowledge; optionally password-protected.
Why local-first
Pet care notes include medication doses, home layout and door codes. Pawsitter runs Gemma 4 26B on the owner’s own machine with llama.cpp’s llama-server and serves the dashboard on the local network only. The whole knowledge base is sent with each question instead of using vector search, so medication instructions are never missed.
Tech: Python, Streamlit, SQLite/SQLModel, and an OpenAI-compatible API to a local llama-server running Gemma.
Built with
AI tools
Prize Categories Entered
- Submitted
Best Use of Gemma 4
Pawsitter runs Gemma 4 26B locally through llama.cpp's llama-server as a focused assistant for pet sitters. It answers only from the owner's notes (medications, feeding, schedules), knows the current time and task status, and shows photos of the right pill bottle or food can. No household data leaves the home network.
- Submitted

Best Open-Source AI Project
The whole AI stack is open: an open-weight model (Gemma 4 26B) served by the open-source llama.cpp server, with a Python/Streamlit app and SQLite. The LLM is the core of the app. It turns the owner's notes into answers a sitter can act on and says 'contact the owner' instead of guessing. The code is public on GitHub.