NEXUS — Meeting-to-Action Intelligence Platform

Turn every enterprise meeting into autonomous executed outcomes.
Not just transcription. Not just summaries. Actual actions taken.

---

The Problem — $37 Trillion of Wasted Work

• 31 hours per month per employee spent in meetings
• 70% of action items from meetings are never completed
• $37 trillion lost annually to poor workplace productivity (Gallup)
• Current tools like Otter.ai and Fireflies only transcribe — they stop at the meeting door

The gap: someone still has to read the transcript, extract tasks, assign owners, create tickets, and send updates. That person is you. Every. Single. Time.

---

The Solution — NEXUS

NEXUS is a multi-agent AI platform that closes the loop automatically:

Meeting audio / transcript
        ↓
[ Voice Agent ]       → Real-time transcription, speaker identification
        ↓
[ Intent Agent ]      → Extracts decisions, action items, domain classification
        ↓
[ Context Agent ]     → Retrieves relevant enterprise knowledge (RAG)
        ↓
[ Domain Expert ]     → Legal / Finance / HR / Ops / Sales / Marketing specialist
        ↓
  High confidence?  ──yes──→ [ Execution Agent ] → Jira / Slack / Linear / Notion
        │ no
        ↓
  [ Human Approval Gate ] → dashboard shows items needing your sign-off
        ↓
[ Live Dashboard ]    → Real-time agent activity, decisions, metrics

One meeting → everything done. No manual follow-up.

---

Why NEXUS is Different

Feature	Otter.ai	Fireflies	Gong	NEXUS
Transcription	✅	✅	✅	✅
Summarisation	✅	✅	✅	✅
Action item detection	⚠ basic	⚠ basic	✅	✅
Domain expert analysis	❌	❌	Sales only	✅ 6 domains
Autonomous execution	❌	❌	❌	✅
Approval gate	❌	❌	❌	✅
Self-hostable	❌	❌	❌	✅
Provider-agnostic LLM	❌	❌	❌	✅
Open source	❌	❌	❌	✅

---

Quick Start — Runs in 5 Minutes

Minimum required: one free API key

# 1. Get a free Groq API key at console.groq.com (takes 30 seconds)
# 2. Clone and run

git clone https://github.com/YOUR_USERNAME/nexus.git
cd nexus
make setup          # installs everything
# Edit .env → add GROQ_API_KEY=your_key_here
make dev            # starts API + dashboard

Open http://localhost:3000 → click Demo → see NEXUS in action.

Or with Docker (no Python setup needed)

git clone https://github.com/YOUR_USERNAME/nexus.git && cd nexus
cp .env.example .env   # add GROQ_API_KEY
docker compose -f docker-compose.free.yml up -d

---

Supported Providers — Change One Line, Switch Everything

# In .env — switch LLM provider without touching any code:

PRIMARY_MODEL=groq/llama-3.3-70b-versatile    # free, fast ← default
PRIMARY_MODEL=gemini/gemini-2.0-flash-exp      # free, Google latest
PRIMARY_MODEL=openai/gpt-4o-mini               # paid, excellent
PRIMARY_MODEL=anthropic/claude-3-5-haiku-20241022  # paid
PRIMARY_MODEL=ollama/llama3.2                  # local, 100% free, no internet

Speech-to-Text:

STT_PROVIDER=groq     # Groq Whisper API (free) ← default
STT_PROVIDER=local    # faster-whisper on CPU (free, no key)
STT_PROVIDER=openai   # OpenAI Whisper API

---

Tech Stack

Layer	Technology	Why
Agent orchestration	LangGraph	Stateful, resumable multi-agent graphs
LLM (any provider)	LiteLLM	One API for 100+ LLM providers
Voice/STT	Groq Whisper / faster-whisper	Free options available
Vector store	ChromaDB (local)	Zero-setup, file-based, portable
Backend	FastAPI + WebSocket	Async, real-time
Frontend	Next.js 14 + Tailwind	Fast, modern
Deployment	Docker + any cloud	AWS, Railway, Render, DigitalOcean

---

Integrations

Execution channels (add as many as you use):

Integration	What it does
Slack	Posts action items to your channels
Jira	Creates tickets with domain-based priority
Linear	Creates issues (great for engineering/product teams)
Notion	Adds tasks to your database
Webhook	Generic — connects to Zapier, Make, n8n, anything

---

Deployment

Local (WSL / Mac / Linux)

make setup && make dev

AWS EC2 Free Tier (12 months free)

# On your EC2 instance (Ubuntu 22.04, t2.micro or t3.small):
REPO_URL=https://github.com/YOUR_USERNAME/nexus.git bash deploy/setup-aws.sh

Railway (free $5/month credit, easiest)

railway init && railway up

Render / Fly.io

Works with the included Dockerfile. Set env vars in their dashboard.

---

API Reference

GET  /api/health          Health check + version
GET  /api/metrics         Live agent performance stats
POST /api/process         Process a meeting transcript
POST /api/upload-audio    Upload audio file → auto-transcribe → process
POST /api/ingest-docs     Add PDFs/DOCX/TXT to knowledge base
POST /api/demo            Run built-in demo (no API key needed for this)
POST /api/approve         Approve a pending action item

WS  /ws/dashboard         Real-time dashboard feed (metrics + results)
WS  /ws/voice/{id}        Stream live audio for real-time transcription

---

Architecture — How the Agents Work

NexusState (LangGraph)
├── extract_intent       → fast LLM: domain, actions, decisions, confidence
├── retrieve_context     → ChromaDB RAG: pull relevant enterprise knowledge
├── domain_agent         → smart LLM: expert analysis per domain
│   ├── LegalAgent       → compliance, contracts, regulatory risk
│   ├── FinanceAgent     → budget, ROI, spend approval
│   ├── HRAgent          → headcount, performance, employment law
│   ├── OpsAgent         → project delivery, vendor management
│   ├── SalesAgent       → pipeline, CRM, deal strategy
│   └── MarketingAgent   → campaigns, content, growth
├── [if confidence ≥ 70%] execute → Slack / Jira / Linear / Notion / Webhook
└── [if confidence < 70%] flag_approval → dashboard approval gate

---

Roadmap

• Multi-tenant SaaS (user accounts, usage limits)
• Stripe billing integration
• Salesforce + HubSpot CRM integration
• Mobile app (React Native)
• Real-time meeting bot (Zoom/Meet/Teams)
• Recurring meeting pattern analysis
• Fine-tuned domain models
• HIPAA compliance mode (healthcare vertical)
• SOC 2 compliance (enterprise sales)

---

Contributing

PRs welcome. See CONTRIBUTING.md.

---

License

MIT — use it, sell it, build on it.