Hacktoberfest 2026 Open Source MIT

🤖 RelayPilot

The AI night-shift engineer for your LLM relay station.
You run the relay. RelayPilot watches it while you sleep.

What it does

• 📡 Channel health monitoring — probes every upstream channel on a schedule: latency, error rate, quota burn.
• 🚨 Automatic key quarantine — an open-weight model classifies key usage locally; a stolen key gets quarantined in minutes, not the morning after.
• 💬 Telegram alerts — three severity levels, quiet hours, and a daily digest. No dashboard spelunking.
• 🔍 Natural-language ops — "which channel burned the most tokens yesterday?" answered in plain language.

Why open?

Key usage logs are the crown jewels of a relay business — they can't leave your server. RelayPilot runs its classifier locally (Ollama + open-weight models, with a built-in heuristic fallback), so the security tool never becomes the security hole. And at thousands of classifications a day, zero marginal inference cost is what makes a $9/month hosted tier actual profit instead of a loss leader.

Self-host in 5 minutes

git clone https://github.com/YOURNAME/relaypilot
cd relaypilot
cp .env.example .env   # fill in your relay + telegram credentials
pip install -r requirements.txt
SIMULATION_MODE=1 ./scripts/demo.sh   # try the key-leak scenario, no real services
# real mode:
SIMULATION_MODE=0 RELAY_API_BASE=https://your-relay.example \
  RELAY_ADMIN_KEY=sk-admin-... uvicorn api:app --host 0.0.0.0 --port 8099

Works with any Veloera / New-API compatible gateway. Simulation mode included for a zero-risk trial.

Pricing

• Self-hosted — free forever (MIT licensed).
• Hosted — $9/month: monitoring, alerting, multi-tenant billing. One prevented quota-burn pays for a year.

Built for the Hacktoberfest 2026 Open-Source AI Challenge — Week 2: Side Hustle.