ethanplusai/jarvis

JARVIS — a voice assistant for Claude Code. Talk to your Mac and he brainstorms a project with you, builds it, and tells you out loud when a Claude Code session needs you. macOS, on your Claude subscription, no API key.

What it solves

JARVIS is a voice-controlled interface for Claude Code, designed to transform the terminal-based AI coding assistant into a hands-free, conversational agent. It solves the problem of managing multiple concurrent AI coding sessions and the friction of manually interacting with CLI tools during complex project design and execution.

How it works

JARVIS acts as a wrapper around the Claude Code CLI. It uses a persistent claude -p process as its "brain," which communicates via WebSockets and FastAPI. The system integrates the Chrome Web Speech API for voice input and Fish Audio for text-to-speech output.

To drive software builds, JARVIS follows a structured pipeline: it brainstorms a design with the user via voice, writes a formal design document to disk, and then spawns a separate Claude Code session to execute the plan using test-driven development. It also monitors all active Claude Code sessions on the machine via AppleScript, notifying the user aloud when a session is stuck or requires human intervention.

Who it’s for

Developers using macOS and Claude Code who want a voice-driven, agentic experience for managing their AI-assisted coding workflows and project builds.

Highlights

  • Voice-Driven Design: Brainstorms project specs out loud and saves them as Markdown files before starting any code execution.
  • Session Monitoring: Watches all Claude Code sessions on the machine and alerts the user when a session is waiting for permission or input.
  • Subscription-Based: Runs on the user's existing Claude subscription rather than requiring a separate API key, with built-in environment scrubbing to prevent accidental API billing.
  • Agentic Build Pipeline: Drives builds through a plan-review-execute cycle, tracking progress via checkboxes in the design spec.
  • MCP Integration: Supports any Model Context Protocol (MCP) server for extended capabilities.
  • Long-term Memory: Maintains a folder of plain Markdown files to remember facts across sessions.

Related

  • Project
    mindsdb/anton

    An open-source self-improving agent that executes real-world tasks like email management and data reporting by writing and running its own code.

  • Project
    ZeroTang05/VideoAdGuard

    A browser extension that uses LLMs to detect and skip sponsored segments in Bilibili videos by analyzing subtitles and audio.

  • Project
    archie0732/healthy-diet-ai-agent

    A Bun and TypeScript-based AI agent for nutrition and healthy diet support featuring RAG, food image analysis, and official health data synchronization.

  • Project
    cloudwego/eino-examples

    A collection of practical examples and demonstrations for the Eino framework, covering AI agents, multi-agent orchestration, and RAG applications.