The risk with natural-language SQL is not a bad query — it is a query against data the asker should not see. Checking workspace permissions on every call, rather than trusting the token issued at connect time, closes that. The narrow surface is a feature too: nothing here can write, and get_data_sources tells the agent what is reachable before it starts guessing table names.
Basedash's public endpoint. The surface is deliberately tiny — one tool to see what data exists, one to ask about it — with their AI analyst generating and validating the SQL behind the scenes.
- ask_question — a plain-English data question, answered with numbers, charts and the reasoning behind the query
- get_data_sources — every connected database, warehouse and SaaS source the user is allowed to see
Browser OAuth on first connection, with a scoped token per client and no API keys anywhere. Workspace permissions are enforced on every call rather than at connection time.
Paste a URL, then authorize — add the endpoint to your client, then approve the OAuth consent screen