Sharing something our team built, in case it is useful to anyone here who works with databases. It is a database client you run on your own server and open in a browser. No desktop install for each person, and your database never has to open a port to the internet because the tool sits inside the same network. MIT licensed, free, no paid tier. One docker command: docker run -p 3000:3000 ghcr.io/libredb/libredb-studio:latest It connects to eighteen database drivers, among them PostgreSQL, MySQL, Oracle, SQL Server, SQLite, MongoDB, Redis, ClickHouse and Cassandra. Two things I want to be straight about rather than oversell: Five of those eighteen are read-only by design, which means you can browse and query them but not write. Those are Druid, Elasticsearch, OpenSearch, Prometheus and Kafka. We list them as read-only rather than quietly counting them as fully supported. There is an AI assistant, but it runs against whatever model endpoint you configure, including a local one. It writes the SQL and never runs it. Running a statement is your own click. We measured that feature instead of claiming it works, over eleven days and 8,199 runs, and the uncomfortable result was that most failures traced back to defects in our own server rather than to the models. We published the protocol and the raw data. Repository: https://github.com/libredb/lib... The measurement write-up: https://arxiv.org/abs/2609.213... If anyone here tries it and something breaks, I would rather hear that than hear nothing.