CatColonyWatch: Local Open AI for Community Cat Field Observation

This article explores the launch of CatColonyWatch, an open-source field-observation tool designed for volunteers and organizations monitoring community cat colonies. The AI system's core concept revolves around the premise that humans, not AI, should observe the cats. The tool streamlines the process by allowing users to focus on observing the animals first, then recording factual notes.

Afterwards, a local open-weight AI model processes these notes, structuring them into a more organized record. Users can then review the AI-generated output before saving or exporting it. The software currently supports various features, such as a reusable database for known cats, individual observations with start and end times, and the ability to calculate durations and total cats observed.

It also captures detailed notes, including food and water observations, feeding-point conditions, biodiversity observations, and visit history. The AI model, powered by Gemma 3 through Ollama, is deliberately designed with narrow boundaries. It doesn't diagnose animals, assess medical urgency, recommend treatments, or infer emotions or intentions.

Its primary function is to help humans organize their observations, not to replace observation itself. The tool is built using Python, Streamlit, SQLite, Ollama, and Gemma 3. It can be accessed and explored through a Streamlit demo available at https://catcolonywatch.streamlit.app/. The AI processing occurs locally on the user's computer, keeping field observations separate from third-party AI servers. This ensures privacy and control over the collected data.