🎨 IdeaForge

Real-time meeting intelligence & visual ideation. Voice → Moodboard in seconds.

IdeaForge listens to your meeting (via browser microphone or text input), extracts key concepts, generates images, and builds a live moodboard — all in real-time.

How It Works

🎤 Voice/Text → 🧠 GPT-4o-mini (concepts) → 🎨 DALL-E 3 (images) → 📋 Live Moodboard

Pipeline:

1. Listener Agent — Transcribes audio (Whisper) and extracts concepts, mood, keywords, color palettes (GPT-4o-mini)
2. Visualizer Agent — Generates images from concepts (DALL-E 3 + Gemini fallback)
3. Curator Agent — Organizes visuals into themed sections (GPT-4o-mini)
4. Presenter Agent — Serves a live moodboard with SSE real-time updates

Quick Start

# Clone
git clone https://github.com/janseninvest/ideaforge.git
cd ideaforge

# Set API keys
cp .env.example .env
# Edit .env with your keys

# Run (zero npm install needed — pure Node.js)
node scripts/demo-session.cjs "My Project" --auto --port 3333

# Open in browser
open http://localhost:3333

Requirements

• Node.js 18+ (uses built-in fetch)
• OpenAI API key — for GPT-4o-mini (concepts) + Whisper (transcription) + DALL-E 3 (images)
• Gemini API key (optional) — fallback image generation

No npm install required — zero external dependencies.

Usage

Browser UI (recommended)

Start the server and open it in your browser:

node scripts/demo-session.cjs "Meeting Title" --auto --port 3333

• 🎤 Record — Click the mic button to record from your browser microphone
• 💬 Type — Click the chat button to type concepts manually
• ⛶ Fullscreen — Press F for TV/presentation mode
• The moodboard updates live as concepts are processed

Programmatic

const { DirectorAgent } = require('./src/agents/director-agent.cjs');

const director = new DirectorAgent();
const session = await director.startSession('My Meeting');
// session.serverUrl → http://localhost:3333

await director.processTextInput('Nordic design with warm wood and soft lighting');
// Board updates automatically via SSE

await director.endSession();

Process Audio Files

node scripts/process-voice.cjs path/to/audio.webm "Session Title"

Environment Variables

# Required
OPENAI_API_KEY=sk-...          # GPT-4o-mini + Whisper + DALL-E 3

# Optional
GEMINI_API_KEY=AI...           # Fallback image generation (Gemini 2.5 Flash)
PORT=3333                      # Server port (default: 3333)

Architecture

src/
├── agents/
│   ├── director-agent.cjs     # Orchestrator — wires all agents together
│   ├── listener-agent.cjs     # Concept extraction (GPT-4o-mini)
│   ├── visualizer-agent.cjs   # Image generation (DALL-E 3 + Gemini)
│   ├── curator-agent.cjs      # Board organization (GPT-4o-mini)
│   └── presenter-agent.cjs    # Board state + server management
├── canvas/
│   ├── server.cjs             # HTTP + SSE server with audio/text endpoints
│   └── public/                # Frontend (vanilla HTML/CSS/JS)
├── data/
│   └── audio-provider.cjs     # Whisper transcription
├── tools/
│   ├── image-generator.cjs    # DALL-E 3 + Gemini image generation
│   ├── palette-generator.cjs  # Color palette extraction
│   └── reference-search.cjs   # Web reference search
└── utils/
    ├── logger.cjs             # Colored console logger
    └── meeting-memory.cjs     # Session transcript storage

Network Access

The server binds to 0.0.0.0 — accessible from any device on the same network. Open http://<your-ip>:3333 on phones, tablets, or TVs.

WSL2 users: You may need to add a port proxy:

# Run in Administrator PowerShell
netsh interface portproxy add v4tov4 listenport=3333 listenaddress=0.0.0.0 connectport=3333 connectaddress=$(wsl hostname -I | cut -d' ' -f1)

Language Support

• Input: Norwegian and English (auto-detected by Whisper)
• Processing: English (all AI calls)
• UI: English

Built With

• OpenAI API — GPT-4o-mini, Whisper, DALL-E 3
• Google Gemini API — Image generation fallback
• Pure Node.js — zero npm dependencies

License

MIT