Personal Jarvis: A Local-First Voice Assistant That Actually Gets Things Done

A new desktop voice assistant called “Personal Jarvis” has been released and is getting attention in developer communities. Unlike typical assistants that just answer questions, Personal Jarvis is designed to actually execute tasks on your machine—running shell commands, driving coding agents, and automating workflows based on your voice requests.

Under the hood, a small router model first estimates how big each job is. Simple requests are handled immediately, while heavier tasks are dispatched to a worker running in an isolated workspace. A separate “critic” process reviews results before they are read back to you. The wake word runs locally via the openWakeWord library, so hotword detection doesn’t depend on any cloud service. Speech recognition can be done fully on-device with Whisper, and conversations are stored as plain Markdown files, making them easy to back up or use with note-taking tools like Obsidian.

Users can choose between running models locally through Ollama or using cloud APIs such as OpenAI and Gemini, with separate settings for the router, worker, and critic tiers. The app already supports Windows, macOS, and Linux on Python 3.11+, and can also run headless on a server that you connect to remotely. For power users and developers who find mainstream assistants too limited, Personal Jarvis aims to be a more capable, automation‑focused alternative.