Inspiration
Powerfly was inspired by a simple question: What if your home had a fitness tracker?
Most people have little visibility into when electricity is cheapest or cleanest to use. While utilities already provide pricing and weather data, acting on that information requires constant attention. We wanted to build an autonomous AI agent that quietly monitors changing conditions and helps homeowners save money without requiring them to constantly check dashboards or graphs.
Instead of using AI for novelty, we wanted to demonstrate how autonomous agents can improve everyday resource efficiency through transparent, trustworthy recommendations.
What it does
Powerfly is an AI-powered home energy assistant that continuously monitors electricity pricing, weather conditions, and household context to recommend the best time to use energy-intensive appliances.
The agent runs on a recurring loop that:
Monitors real-time electricity prices and local weather Understands household context such as home orientation and appliance usage Detects opportunities to reduce electricity costs Sends proactive recommendations like: "Electricity is 50% cheaper right now. Run the dishwasher." "Pre-cool your home this morning before peak pricing begins."
Unlike traditional dashboards, Powerfly explains why it made each recommendation through a transparent Trust Ledger, allowing users to see how changing conditions influenced the agent's decisions.
How we built it
Powerfly was designed as a mobile-first web application centered around an autonomous agent loop.
Our system periodically gathers information from multiple sources, including:
Real-time electricity pricing Local weather forecasts Home characteristics and orientation Utility usage history Appliance profiles
To make continuous inference economically feasible, we explored lightweight local models for routine reasoning while reserving larger models only when needed.
The agent continuously:
Collects new data Evaluates energy-saving opportunities Generates personalized recommendations Re-evaluates its decisions as conditions change Updates its recommendations through the Trust Ledger
Built an autonomous Observe → Reflect → Plan → Act agent loop using FastAPI and Python. Integrated live CAISO electricity pricing and Open-Meteo weather to generate real-time energy recommendations. Used Claude (Haiku/Sonnet) to rank personalized recommendations based on household context. Implemented a self-correcting feedback loop that learns from user interactions and suppresses ignored recommendations. Scheduled recurring execution with APScheduler and exposed REST APIs for the frontend, notifications, and rewards system. Stored household profiles, snapshots, nudges, and user patterns in SQLite, with deployment support for Akash and integrations with Nexla and Zero.xyz.
Challenges we ran into
One of our biggest challenges was balancing intelligence with practicality.
Reliable real-time electricity data is much easier to obtain than water or gas usage, so we narrowed our MVP to electricity instead of trying to solve every utility problem at once.
Because the system continuously loops, inference costs also became a major consideration. Rather than relying entirely on large cloud models, we explored lightweight local models that could perform routine reasoning more efficiently.
Finding trustworthy regional electricity pricing sources also required more research than expected because pricing varies widely between utility providers and locations.
Accomplishments that we're proud of
We are especially proud that Powerfly goes beyond simply displaying energy data.
Instead, it acts as a true autonomous assistant that continuously observes, reasons, and adapts to changing conditions.
Highlights include:
A continuously running agentic loop that proactively recommends energy-saving actions A warm, approachable interface that makes complex energy information easy to understand The Trust Ledger, which transparently explains why the AI made or changed its recommendations A product experience that transforms complicated energy optimization into simple everyday habits What we learned
Building Powerfly taught us that energy efficiency is often less about using less electricity and more about using it at the right time.
We also learned that effective autonomous agents do not always require massive models. By combining smaller reasoning models with reliable external data, it is possible to build practical, cost-effective AI systems that continuously adapt to changing environments.
Most importantly, we learned that users are much more likely to trust autonomous systems when those systems clearly explain their reasoning instead of behaving like a black box.
What's next for Powerfly AI
We are excited to expand Powerfly beyond electricity into a complete home resource management platform.
Future plans include:
Integrating water and natural gas monitoring Connecting with smart home devices to automatically shift appliance usage during optimal times Supporting additional utility providers and regional pricing markets Partnering with utility companies and smart home platforms to improve grid efficiency while helping households reduce their energy costs
Our long-term vision is for Powerfly to become a trusted AI energy manager that quietly works in the background, helping every home become smarter, more efficient, and more sustainable.
Log in or sign up for Devpost to join the conversation.