jarvis

Jarvis is a private, voice-first assistant that runs on your computer. The README says speech recognition, language models and speech synthesis run on hardware you control, and that your conversation and data stay on the computer. It is designed to take part in a conversation: while listening it keeps a short, temporary rolling transcript of nearby speech so that "Jarvis, what do you think?" can use what was just discussed. You can address it by name anywhere in a sentence.

Listed capabilities: memory across sessions (a local diary and knowledge graph with a Memory Viewer), built-in tools for web search, weather, time, screenshot OCR, file access, nutrition tracking and optional location, MCP servers for browser automation and smart-home control, push-to-talk dictation into other apps, and a companion text chat with silent replies. The README says sensitive information is redacted before it reaches model context or the saved diary, and that web search, weather and connected tools use the network only when asked.

Download a build for macOS (Apple Silicon or Intel), Windows x64 or Linux x64 from GitHub Releases. The setup wizard walks through speech recognition and a model server: Ollama, or an OpenAI-compatible server such as LM Studio or llama.cpp. Suggested chat models range from `qwen3.5:0.8b` on small hardware to `gemma4:e4b` by default and a larger local setup, with a memory budget shown during setup. The README says it is developed mainly on macOS, that Windows and Linux behavior may differ and that accuracy depends on your models; it links evaluation results. GitHub did not detect an SPDX license, so check the LICENSE file. Last push was 2026-10-13. This review is based on the README and repository data; we have not run it ourselves.