HackEurope · Sustainability Track

Break down any task into subtasks, route each to the smallest capable model, run them in parallel, and track cost and CO₂ savings in real time. Built with a Next.js frontend and a Python (FastAPI) backend, using Ollama Cloud or local Ollama for inference, plus optional Stripe wallet top-ups for per-agent billing.

• Task decomposition — A single prompt is decomposed into subtasks; each subtask is assigned a model (by parameter size/capability) and optional dependencies.
• Parallel execution — Agents run concurrently where the dependency graph allows; progress streams to the UI (graph view, timeline, live output).
• Carbon tracking — Pipeline CO₂ (gCO₂) is estimated from token counts and GPU energy models; compared to a 70B single-model baseline. Real-time grid intensity (France by default) from Electricity Maps or a built-in fallback.
• Green window scheduling — Optional “run when grid is cleaner” using 24h history + 8h forecast (when Electricity Maps key is set).
• Billing — Demo wallet (SQLite ledger); optional Stripe integration for top-ups and payment-method storage. Backend seeds the demo user with $15 on first run.

• Node.js 18+ and npm
• Python 3.10+
• Ollama — ollama.com: either local install or an Ollama Cloud API key
• Stripe (optional, for real top-ups): account + Stripe CLI for local webhooks

Clone the repo. Create a .env file in the project root (same folder as backend/ and frontend/). Backend and frontend read from this file.

Required for basic run:

NEXT_PUBLIC_API_URL=http://localhost:8000

For Ollama Cloud (recommended):

OLLAMA_API_KEY=your_ollama_api_key

For Stripe top-ups (optional):

STRIPE_SECRET_KEY=sk_test_...
STRIPE_WEBHOOK_SECRET=whsec_...
NEXT_PUBLIC_STRIPE_PUBLISHABLE_KEY=pk_test_...

For live carbon intensity (optional):

ELECTRICITY_MAPS_API_KEY=your_em_key

Without ELECTRICITY_MAPS_API_KEY, the app uses a France fallback intensity (~65 gCO₂/kWh).

cd backend
python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --reload --host 0.0.0.0 --port 8000

• Ollama Cloud: set OLLAMA_API_KEY in .env; no local Ollama needed.
• Local Ollama only: leave OLLAMA_API_KEY unset, run Ollama locally, and pull models (e.g. ollama pull llama3.2).

On first run, the backend seeds the demo user with $15 so you can run agents without Stripe. Health: http://localhost:8000/api/health.

cd frontend
npm install
npm run dev

Open http://localhost:3000.

Stripe cannot reach localhost. Use the Stripe CLI to forward events:

./scripts/setup-stripe.sh            # one-time: download CLI + log in
./scripts/setup-stripe.sh --listen   # forward webhooks (run in a separate terminal)

Use the printed whsec_... as STRIPE_WEBHOOK_SECRET in .env. To only print the secret: ./scripts/setup-stripe.sh --print-secret.

• Backend seeds demo with $15 on first start.
• To add more without Stripe, call the demo top-up API:

curl -X POST http://localhost:8000/api/billing/topup \
  -H "Content-Type: application/json" \
  -d '{"user_id":"demo","amount_usd":5}'

• Set STRIPE_SECRET_KEY, STRIPE_WEBHOOK_SECRET, and NEXT_PUBLIC_STRIPE_PUBLISHABLE_KEY in .env (use sk_live_... / pk_live_... for production).
• In Stripe Dashboard → Developers → Webhooks, add an endpoint: https://YOUR_BACKEND_DOMAIN/api/stripe/webhook.
• Events: payment_intent.succeeded, payment_intent.payment_failed.
• Wallet credits occur only for payment_intent.succeeded with metadata purpose=wallet_topup and user_id; the app sets these when creating top-ups via POST /api/billing/create_topup_intent.

• Frontend: Next.js 15, React 19, Tailwind, React Flow (graph), Stripe React/JS.
• Backend: FastAPI, Ollama (cloud or local), SQLite (billing ledger + user→Stripe mapping), Stripe, Electricity Maps (optional).
• Carbon: Energy model from GPU TDP/throughput; carbon intensity from Electricity Maps or zone fallbacks; 70B baseline comparison and green-window suggestion.