Live ยท MIT License

Metastrome.ai

The Multiverse of You

An AI-powered brainstorming room where you talk to expert personas โ€” all variants of the same person from parallel universes. Think Google Meet, but your colleagues are AI agents with real backstories, opinions, and expertise.

โŸฉ Your Expert Variants

8 personas, each with a unique career path. Fully configurable via JSON.

๐Ÿš€

The Visionary

Startups, fundraising, go-to-market

๐Ÿ—๏ธ

The Architect

Distributed systems, API design, security

โšก

The Builder

Full-stack dev, React, testing, DevOps

๐Ÿงฌ

The Scientist

Deep learning, NLP, causal inference

โš™๏ธ

The Machinist

MLOps, GPU optimization, model serving

๐Ÿ“Š

The Datasmith

Data modeling, dbt, Spark, pipelines

๐ŸŽฏ

The Strategist

Product strategy, user research

๐ŸŽจ

The Artist

Interaction design, accessibility

โŸฉ What It Does

A full-stack real-time AI meeting experience.

Runs on Your Machine

Windows, macOS, or Linux. No cloud required โ€” self-hosted for full privacy.

Any Chat App

Talk via WhatsApp, Telegram, Discord, Slack, Signal, or iMessage.

Persistent Memory

ChromaDB dual-layer vector memory โ€” shared + per-agent. Remembers across sessions.

Emotion Detection

Webcam frames analyzed via Gemini Flash every 5s. Agents adapt their tone.

Real-time WebRTC

LiveKit server for ultra-low-latency audio/video with echo cancellation.

Multiple TTS Providers

Edge (free), Deepgram, Cartesia, or ElevenLabs โ€” switch in .env.

โŸฉ How It Works

1.

You join a meeting room and pick which variants you want.

2.

You speak or type your idea โ€” voice goes through LiveKit WebRTC โ†’ Deepgram STT.

3.

Your emotion is detected from your webcam every 5 seconds via Gemini Flash.

4.

The 3-phase conversation orchestrator plans the response: Plan โ†’ Primary โ†’ Follow-up.

5.

Agents react to each other โ€” agreeing, pushing back, building on points.

6.

Agents can [pass] if they have nothing meaningful to add.

7.

TTS converts each response to audio โ€” sentences streamed in parallel.

8.

You can interrupt anytime โ€” speaking mid-agent cancels their audio immediately.

โŸฉ Tech Stack

Backend

AutoGen (multi-agent)
FastAPI + Uvicorn
LiveKit Server SDK
Edge TTS (free default)
ChromaDB vector memory

Frontend

Next.js 16 + React 19
Tailwind CSS v4
livekit-client (WebRTC)
TypeScript 5

Cloud Services

OpenRouter โ€” LLM gateway
Deepgram โ€” STT + TTS
Gemini Flash โ€” emotion
LiveKit โ€” WebRTC SFU

Brainstorm with your variants.

No GPU required. Clone the repo, add your OpenRouter key, and start a meeting in minutes.

Get Started on GitHub
Panaiq - Strategic AI Transformation Partner | Custom AI Solutions & Automation