Tesla Powerwall scheduler
I recently built an app using Replit, Teslemetry, and Resend, and overall it was a pretty interesting experience.
The primary goal of the app is to automatically charge my two Tesla Powerwall 2 batteries during off-peak electricity hours, but only when it makes sense.
The app checks whether the current solar irradiance is below 150 W/m² and also looks at the next day's weather forecast. If the forecast predicts poor solar production—such as heavy cloud cover, rain, or snow—it charges the Powerwalls overnight when electricity is cheaper.
One thing I really liked about Replit was how easy it was to describe what I wanted to the AI agent. It did a good job generating the initial application, and once I had a working version, I even asked it for ideas on how to improve it. It came up with a lot of useful suggestions that I hadn't considered.
The biggest challenge came during debugging. I spent a lot of time getting the AI agent to fix features that either weren't working correctly or weren't behaving the way I expected. Those back-and-forth iterations consumed credits pretty quickly. In the end, I spent close to $100 in AI credits.
I'm still experimenting with Home Assistant Green to accomplish the same automation. My hope is that, once everything is configured, it will provide similar functionality with a much lower ongoing operating cost than relying heavily on AI-assisted development.
Overall, I'm glad I built the app because I learned a lot, but if you're planning a similar project, it's worth budgeting for AI credits—especially if your project requires a lot of troubleshooting and refinement.