AI Mix Masters
(formerly Trap Master Reference / Master Mix Workbench — the GitHub repo slug stays the same; only the user-facing name has changed.)
A mixing and mastering workbench — live genre-aware plugin picks, meter targets, a troubleshooter, and a voice-chat mastering assistant powered by OpenAI's Realtime API and (optionally) Claude with live web search for niche knowledge lookups.
Voice migration in flight: the voice assistant is being moved from OpenAI Realtime to ElevenLabs Conversational AI (Claude Sonnet 4.6, Hope voice, Expressive Mode) on the voice-elevenlabs branch. The OpenAI path on main stays the production fallback until the new path is solid.
🔗 Live demo: https://begb0037admin.github.io/aimm/
Built as a single-file HTML app. No build step, no backend. Your state (chains, notes, favorites, API keys) lives in your browser's localStorage.
Quick start (local)
python3 -m http.server 8000
then open http://localhost:8000 in Chrome or Safari.
Or use the hosted version on GitHub Pages — link at the top of this repo's description once it's enabled.
Voice chat setup
- Sign up at https://platform.openai.com and add a few dollars of pay-as-you-go credit.
- Create an API key at https://platform.openai.com/api-keys — starts with
sk-.
- In the app, open the Voice Chat tab, paste the key, hit Save.
- Hit the big Start call button. The browser will ask for mic permission once.
The key is stored only in your browser. It's never uploaded anywhere except directly to OpenAI when you start a call.
Default model is gpt-4o-mini-realtime-preview (cheap but a bit dumb). Switch the Voice model dropdown to gpt-4o-realtime-preview for better accuracy — roughly 10× the cost per minute, still pennies for a short session.
Optional: Anthropic key for the research tool
The voice model is fast and conversational but its knowledge of niche producer techniques (specific engineers, plugin parameters, current 2024–2026 trends) is shaky and it tends to bluff. To fix that, paste a Claude API key (sk-ant-…) into the second key field. When the voice AI is about to bluff on something niche, it'll instead call Claude with live web search, get a grounded answer with citations, and read it back to you. Adds ~2–4s of "hold on, let me check…" but kills the hallucinations.
Get a key at https://console.anthropic.com → Settings → API Keys → Create. ~$5 in credits is plenty to start. Pick the research model in the dropdown — Sonnet 4.6 is the balanced default; Opus 4.6 if you want maximum smart.
What's in it
- 450+ plugin library: Waves, iZotope, UA, Softube, SSL Native, Plugin Alliance, Native Instruments, Sonnox, Slate, Eiosis, oeksound, Soundtoys, Antares, Celemony, LiquidSonics, and more.
- Genre presets: trap, drill, UK drill, phonk, rage, cloud, Memphis, Jersey club, hyperpop, boom bap.
- Per-bus chains (master / vocal / 808 / drums / FX) with inline settings notes that stick to each plugin.
- Platform loudness targets: club-loud, Spotify, Apple Music, SoundCloud.
- Troubleshooter: flag symptoms (harsh hats, muddy 808, pumping, etc.), get recipe chains.
- Voice AI that can read your workbench state, add/remove plugins, switch genre, flag issues, and pin settings to plugins via function calling.
- Deep research (with an Anthropic key): voice model delegates niche / knowledge-heavy questions to Claude with live web search instead of bluffing.
Collaboration
Every user has their own local state. To share work:
- Snapshot (top right) copies the whole workbench as markdown — paste into Slack, email, Notion.
- Import/export of session state is planned.
For live collaboration (multiple people on the same workbench at once) a small backend would be needed — not there yet.
Privacy
- API key: browser localStorage only.
- Voice audio: browser ↔ OpenAI direct, via WebRTC.
- State: browser localStorage only.
- Nothing is sent to any server run by the repo author.
Stack
- Single-file HTML + vanilla JS + CSS. No frameworks.
- OpenAI Realtime API (WebRTC + data channel) for voice.
- 13 function-calling tools expose workbench state to the model.
License
Private / personal — add your own license if you fork it.
