ccvdemo

Example Claude Code project that uses ccv for realtime voice.

Install

1. Launch this directory with npx -y --package github:AnEntrypoint/ccv ccv-claude. That wrapper starts Claude Code with --dangerously-load-development-channels server:ccv, which is what makes speech arrive on its own; plain claude still works, but the session has to call voice_listen to be woken by speech. Either way .mcp.json loads and starts npx -y github:AnEntrypoint/ccv.
2. Optional: copy .env.example to .env and set HF_TOKEN for faster model downloads.
3. Ask Claude to run voice_setup, then voice_start, then open http://127.0.0.1:8800/ and click join.
4. For the chatjimmy scripts (start.mjs, chatjimmy-bridge.mjs), run npm install in this directory once.

Models and LiveKit install into .ccv/ in this directory (git-ignored).

Discord

Set DISCORD_TOKEN and DISCORD_CHANNEL (a voice channel URL or ID) in .env, then start Claude Code from this directory with claude --dangerously-load-development-channels server:discord. Claude Code launches the discord MCP server from .mcp.json, which joins the voice channel, transcribes speech with ccv's models, and delivers each utterance to the session. Replies go out with discord_speak. Without the flag, call discord_listen to wait for speech.

WebRig runner

webrig-runner/ is the Discord-free counterpart of discord-runner/. Instead of joining a Discord voice channel itself, it connects to a running WebRig audio bridge over WebRig's audio WebSocket, so the WebRig instance keeps owning the page capture and the voice connection.

1. Start WebRig with its Discord bot configured (DISCORD_BOT_TOKEN, GUILD_ID, CHANNEL_ID) and its audio WebSocket listening (WS_AUDIO_PORT, default 9888). The browser extensions are audio clients: in browser-extension mode the socket is served by node companion/index.js (npm run companion), not by the extension itself.
2. Run npm install inside webrig-runner/ once, then node webrig-runner/index.mjs (or npm start there). It loads the same ccv speech models, follows .webrig/heard.jsonl, and starts chatjimmy-bridge.mjs against it.
3. Replies are synthesized with ccv's TTS and sent back to WebRig as stereo Float32 AUDIO frames; WebRig plays them into the voice channel.

Configuration is the WEBRIG_* block in .env.example. Set WEBRIG_BRIDGE=0 to run only the audio bridge without the chatjimmy responder.