For years I kept building admin dashboards, at work and for my own projects. Every one of them was the same thing: a UI wrapped around a growing pile of SQL queries. Then requirements would change, the schema would change, and the dashboard was always a week or two behind reality. At some point I just stopped building them and wrote the queries directly when I needed something.
That worked until the queries got complex enough that I was losing half my day in a SQL editor. So I tried ChatGPT and Claude. The loop was: paste the schema, explain the relationships, ask the question, get a half-right query because the model never had the full picture. Then do it again for the next question. Not sustainable, especially once the team needed the same answers.
So I built a free tool: alodb. It reads your schema once, then you just ask questions in a chat. Setup takes about 5 minutes. Needs a Gemini API key for now, Claude and OpenAI are next on the list.
Two things I cared about from day one:
- Your DB credentials and row data never leave your machine. The agent only ever sees schema structure. Every query is executed locally on your side, not on our server. Works fine with databases behind a firewall for the same reason. - The backend agent is fully open source, so you can read exactly what it does with your schema: github.com/mololab/alodb
macOS only right now (Windows and Linux coming). Supports standard PostgreSQL and Postgres-compatible hosts like Supabase.
Happy to answer anything. I'll stick around for the next few hours.