TerraGemma
An offline-first wilderness exploration AI companion based on Google's open-source Gemma 3 model, designed to run 100% locally. It supports sensory navigation, wilderness first-aid triage, and plant and animal identification, targeting outdoor scenarios without network connectivity.
Timeline (last 90 days)
TerraGemma is an offline-first wilderness adventure companion powered by Google's open-weight model Gemma 3 that runs 100% locally.
TerraGemma is built with Python and Streamlit, and is designed for offline execution on consumer-grade hardware.
TerraGemma primarily uses the gemma3:1b model, which is approximately 815 MB and completes inference in approximately 500 ms on an ordinary laptop CPU.
TerraGemma calls gemma3:4b for deeper reasoning on complex wilderness first aid and plant identification.
TerraGemma's models connect via the Ollama local HTTP API, with keep_alive set to 60m to keep the weights resident in memory.
TerraGemma implements the NOAA solar algorithm in pure Python, using the official sunset zenith angle of 90.833° and civil twilight angle of 96°.
TerraGemma synthesizes speech locally in the browser via a self-contained HTML/JS component that calls window.speechSynthesis (the Web Speech API), without invoking any external cloud-based TTS.
TerraGemma uses folium and streamlit-folium to render OpenStreetMap layers.
TerraGemma separates deterministic computation (solar, turnaround, maps) from generative reasoning (sensory landmarks, first aid, identification); the former has zero dependencies, while the latter is handled by a local LLM with a heuristic fallback engine.
All Facts (9)
TerraGemma is an offline-first wilderness adventure companion powered by Google's open-weight model Gemma 3 that runs 100% locally.
50%UnverifiedTerraGemma is built with Python and Streamlit, and is designed for offline execution on consumer-grade hardware.
50%UnverifiedTerraGemma primarily uses the gemma3:1b model, which is approximately 815 MB and completes inference in approximately 500 ms on an ordinary laptop CPU.
50%UnverifiedTerraGemma calls gemma3:4b for deeper reasoning on complex wilderness first aid and plant identification.
50%UnverifiedTerraGemma's models connect via the Ollama local HTTP API, with keep_alive set to 60m to keep the weights resident in memory.
50%UnverifiedTerraGemma implements the NOAA solar algorithm in pure Python, using the official sunset zenith angle of 90.833° and civil twilight angle of 96°.
50%UnverifiedTerraGemma synthesizes speech locally in the browser via a self-contained HTML/JS component that calls window.speechSynthesis (the Web Speech API), without invoking any external cloud-based TTS.
50%UnverifiedTerraGemma uses folium and streamlit-folium to render OpenStreetMap layers.
50%UnverifiedTerraGemma separates deterministic computation (solar, turnaround, maps) from generative reasoning (sensory landmarks, first aid, identification); the former has zero dependencies, while the latter is handled by a local LLM with a heuristic fallback engine.
50%