Jarvis

Jarvis is a local-first voice assistant that runs as a desktop app on macOS, Windows, and Linux. Speech recognition, the language model, and speech synthesis all run on hardware you control, and no cloud AI service is required. You address Jarvis by name anywhere in a sentence and follow up in natural speech.

It is built to join a conversation as a third participant. While listening, it keeps a short, temporary rolling transcript of nearby speech, so a question such as "Jarvis, what do you think?" is answered in the context of what was just discussed rather than as an isolated sentence. A companion chat window shares the same conversation and memory for silent text replies.

Models run through Ollama or any OpenAI-compatible local server, such as LM Studio, oMLX, or llama.cpp. A setup wizard picks speech recognition and chat models and shows a memory budget, with suggested chat models ranging from small Qwen models to Gemma variants, plus a separate fast model for voice intent and tool routing. Whisper handles speech recognition, and Piper (or Chatterbox from source) handles speech output.

Conversations are kept in a local diary and knowledge graph that you can browse in a Memory Viewer, and sensitive information is redacted before it reaches model context or the diary. Built-in tools cover web search, weather, time, screenshot OCR, file access within the home directory, nutrition tracking, and optional location, and MCP servers add browser automation, smart-home control, and other integrations. Jarvis is developed primarily on macOS, and accuracy depends on the chosen models and hardware.