A developer released Larmor, an open-source (MIT) tool adding real two-way voice interaction to coding agents like Claude Code. Building it required solving echo cancellation, interruption detection, and mid-task listening, which the author says most voice-agent setups skip. It runs fully locally on Apple Silicon Macs using local speech models and works with any MCP-supporting agent.
live-vibe is a voice plugin for Claude Code that keeps the microphone open full-duplex: it answers when you finish a thought and stops when you talk over it. It is built as a Claude Code Mod (plugin) that runs code in the agent's path and draws into the terminal, using CUDA and Apple silicon for voice processing. It targets developers who want ChatGPT Voice Mode-style hands-free conversation with Claude Code.
live-vibe is a voice plugin for Claude Code that keeps the microphone open full-duplex: it answers when you finish a thought and stops when you talk over it. It is built as a Claude Code Mod (plugin) that runs code in the agent's path and draws into the terminal, using CUDA and Apple silicon for voice processing. It targets developers who want ChatGPT Voice Mode-style hands-free conversation with Claude Code.
Banshee lets Claude Code speak questions aloud and listen to spoken answers via three MCP tools, so developers can step away from the desk. It uses Whisper for listening and Kokoro for speech, all running locally on a Mac. It solves the problem of Claude Code idling while waiting on a one-word reply.
Banshee lets Claude Code speak questions aloud and listen to spoken answers via three MCP tools, so developers can step away from the desk. It uses Whisper for listening and Kokoro for speech, all running locally on a Mac. It solves the problem of Claude Code idling while waiting on a one-word reply.
claude-speak is an open-source plugin that gives Claude Code voice output, reading responses aloud using the OS's built-in text-to-speech (say on macOS, espeak-ng on Linux). It requires no API keys or cloud calls, and can read responses on demand or automatically after each turn. It solves the problem of having to read responses off the screen after using voice input.
claude-speak is an open-source plugin that gives Claude Code voice output, reading responses aloud using the OS's built-in text-to-speech (say on macOS, espeak-ng on Linux). It requires no API keys or cloud calls, and can read responses on demand or automatically after each turn. It solves the problem of having to read responses off the screen after using voice input.
A developer on the SKI team (heyski.io) posted to Reddit asking whether anyone uses voice coding with AI agents, disclosing their involvement. SKI is a free, on-device tool that lets users hold a key, speak to agents like Claude Code, Codex, and Cursor, and hear responses aloud. The post seeks community feedback on voice-based coding workflows.
A developer on the SKI team (heyski.io) posted to Reddit asking whether anyone uses voice coding with AI agents, disclosing their involvement. SKI is a free, on-device tool that lets users hold a key, speak to agents like Claude Code, Codex, and Cursor, and hear responses aloud. The post seeks community feedback on voice-based coding workflows.
proofcut is a video editor driven by an AI agent via MCP (e.g. Claude Code) that edits talking-head videos by transcript: cutting, adding b-roll, music, captions, and rendering. It self-verifies by transcribing the rendered output and comparing words and frame count against the edit. Everything runs locally on the user's machine, solving slow timeline-based editing for transcript-first editors.
proofcut is a video editor driven by an AI agent via MCP (e.g. Claude Code) that edits talking-head videos by transcript: cutting, adding b-roll, music, captions, and rendering. It self-verifies by transcribing the rendered output and comparing words and frame count against the edit. Everything runs locally on the user's machine, solving slow timeline-based editing for transcript-first editors.