Inspiration
Underneath all of AI and robotics is one thing: electricity. Every model, data center and robot runs on it. Soon people and companies won't just use power, they'll store it, buy it and sell it, like they do with housing. We built the AI that trades it.
What it does
Ask in plain English or by voice: "How much will my battery make tomorrow?", "keep 30% for the storm", "approve." AI agents plan every buy and sell on real New York grid prices, check it in a sandbox, and wait for your approval.
On real prices for a 10 MW battery in NYC: $419 tomorrow vs $186 for a fixed schedule, and $14,055 vs $6,081 over the last 30 days.
How we built it
- AWS Bedrock (Claude) + deepagents: an analyst, a trader and a risk officer who can veto
- Optimizer: the math picks the hours, not the AI
- Daytona: every plan is verified in a sandbox; the backtest runs 30 sandboxes in parallel
- ElevenLabs: talk to the trader
- Anchor: reads live grid alerts from NYISO's site
- SurrealDB: remembers your battery and rules
- Arize: checks every number the AI says
- Hatchet, OpenUI, AgentGlow: scheduling, dynamic answers, live backend view
Challenges
Keeping the AI's numbers honest, and keeping sandboxes and browser sessions alive and cleaned up.
What we learned
AI agents are most trustworthy when math and a sandbox enforce the rules.
Built With
- anchor-browser
- arize
- aws-bedrock
- claude
- daytona
- deepagents
- docker
- elevenlabs
- fastapi
- hatchet
- langchain
- nyiso
- openui
- python
- react
- surrealdb
- typescript
Log in or sign up for Devpost to join the conversation.