Gemma.Witness - Offline Multimodal Evidence Capture with Gemma 4
Gemma.Witness, submitted to the Gemma 4 Challenge, is an offline-first multimodal evidence capture system designed for environments where cloud access, trust, or chain-of-custody assumptions fail. Built with Rust, Tauri, and local inference orchestration, it records audio alongside supporting images and runs local multimodal analysis through Gemma 4. The system produces a signed evidence bundle containing structured incident reports, timestamped evidence metadata, local reasoning traces, hash-linked verification artifacts, and exportable forensic bundles. The focus is on reliability and local verification rather than AI assistant behavior. The application runs fully local through a desktop interface. Source code is available on GitHub at https://github.com/moonrunnerkc/gemma-witness. Gemma 4 serves as the reasoning layer behind the entire evidence pipeline.
Enables developers to build trustworthy evidence capture systems that work offline and without centralized APIs.