dsh-dbhub-live

Let an AI model write SQL, inspect schemas, and pull ad-hoc data for you without ever seeing a single credential byte. dsh-dbhub-live redraws the trust boundary between DeepSeek Harness and your databases: the model only references a workspace name, an environment label, and public metadata (engine, host, port, database). Passwords, user names, and full DSN strings stay entirely on the host side, entered through the Web settings panel, so nothing sensitive flows through the model context.

Every SQL execution spawns a short-lived dbhub process that is killed the moment the query finishes. There is no long-running service, no shared port, and no lock file, so parallel tasks and multiple DSH instances on the same machine simply coexist. The tool surface is fixed at four declarations (configure, list sources, execute SQL, search objects) regardless of how many environments you manage, keeping the model context lean. Supported engines include MySQL, PostgreSQL, MariaDB, SQLite, and SQL Server. Connections are isolated per workspace and per environment, auto-discovered from mise env and .env files, and dbhub itself is installed on first use and kept up to date in the background.

It fits engineers who use DSH for everyday development work and want the AI to query databases, review schemas, or pull data, especially teams juggling dev, test, and prod environments. The browser status card offers add, edit, test, and toggle controls for connections at a glance. When authentication fails, the tool points you back to the settings UI to update credentials, and the model is never asked for a password, keeping the security boundary clean and obvious.