Jarvis is a 100% private, offline AI voice assistant designed to run locally on your computer. It allows for natural, conversational interaction, acting as a third person in the room, and offers extensive capabilities including memory, web access, application control, and integration with numerous tools. Its primary benefit is providing a powerful, privacy-focused AI assistant without cloud reliance, subscriptions, or data harvesting, automatically redacting sensitive information before saving.

How It Works

Jarvis operates entirely locally, leveraging open-source models for speech recognition (Whisper), text-to-speech (Piper TTS), and large language model (LLM) inference via Ollama. Its core design emphasizes conversational awareness, maintaining short-term context and understanding wake words placed anywhere in a sentence. Unlimited memory is achieved by storing and indexing past interactions, allowing for context-aware responses and adaptive tone. Extensibility is handled through the Model Context Protocol (MCP), enabling integration with thousands of external tools and services via smart, embedding-based tool selection that prevents performance degradation with more integrations.

Quick Start & Requirements

• Primary Install: Download the appropriate release zip/tar.gz for your OS (Windows, macOS, Linux) from GitHub Releases, extract, and run the executable.
• Prerequisites: Ollama must be installed. Windows users may need the Visual C++ Redistributable.
• Hardware: Minimum 8GB VRAM recommended for the default gemma4:e2b model; 16GB+ for better quality models like gemma4:e4b, and 24GB+ for high-end models like gpt-oss:20b. VRAM requirements include the intent judge model.
• Setup: A setup wizard guides through model selection and installation. A free MaxMind GeoLite2 account and database download are required for location detection.

Highlighted Details

• Privacy-First: 100% local processing, no cloud services, no data harvesting, and automatic redaction of sensitive information (emails, passwords, tokens).
• Conversational AI: Understands context within ongoing discussions, allowing natural follow-ups without repeating commands. Wake word ("Jarvis") can be spoken anywhere in a sentence.
• Unlimited Memory: Stores and indexes all conversations, adapting tone and learning user preferences over time. Includes a Memory Viewer GUI.
• Offline Dictation: A free, private alternative to cloud dictation services, allowing users to speak and paste text into any application via a hotkey.
• Extensible Tooling (MCP): Integrates with thousands of services (Home Assistant, Google Workspace, GitHub, Notion, Slack, databases) via MCP servers.

Maintenance & Community

Primary development is focused on macOS, with Windows and Linux support potentially lagging. The project encourages open contributions and issue reporting. Specific community links (Discord, Slack) or a public roadmap are not detailed in the README.

Licensing & Compatibility

Jarvis is free for personal use. Commercial use requires contacting the developers, indicating a custom or proprietary licensing model for business applications rather than a standard open-source license.

Limitations & Caveats

The project is under active development, with potential for Windows/Linux support to trail macOS. Currently, it features a voice-only interface and lacks mobile applications. Known issues include occasional filtering of "stop" commands as echo and dictation incompatibility on macOS 26+ due to pynput. Location detection requires manual setup of a GeoLite2 database.