Everett_DM

AI Dungeon Master. Everett's cloned voice narrates. NPCs speak in their own synthetic voices. RAG over campaign lore and D&D rules. Real-time Discord voice integration.

See CLAUDE.md for the full identity and architecture doc. See roadmap.md for the phased build plan.

---

Current State

Working voice pipeline inherited from the josef-voice project:

You speak → Discord (Opus 48k) → VAD on silence
          → ElevenLabs Scribe (STT) → transcript
          → Claude brain (persona, context)
          → ElevenLabs Flash v2.5 (streaming TTS) → Opus → speaks in channel

• Auto-follows you into voice channels
• Real-time STT + TTS, latest-utterance-wins (no stale reply queue)
• Persona lives in persona.md — edit to retune without touching code
• Voice ID in voice-config.json

Setup

cd C:\Users\derek\Documents\GitHub\Everett_DM
npm install
npm run preflight   # checks token, ElevenLabs key, voice id, ffmpeg
npm start

.env (create in this folder)

DISCORD_BOT_TOKEN=...
ELEVEN_API_KEY=...
OWNER_USER_ID=...        # optional: only respond to this Discord user ID
VOICE_GUILD_ID=...       # optional: restrict to one server
BRAIN_MODEL=claude-haiku-4-5   # or claude-sonnet-4-6 for deeper answers

Files

src/index.js      orchestrator: auto-follow, per-turn pipeline
src/voice.js      Discord receive (Opus→WAV) + playback
src/stt.js        ElevenLabs Scribe speech-to-text
src/tts.js        ElevenLabs Flash streaming TTS
src/brain.js      Claude brain (swappable)
src/config.js     config + system prompt builder
src/log.js        logger
preflight.js      readiness check
persona.md        DM persona / system prompt (edit this to retune)
voice-config.json ElevenLabs voice ID + model
CLAUDE.md         project identity and architecture
roadmap.md        phased build plan