The failure mode of natural-language analytics is a query that runs, returns a number, and means something different from what was asked — wrong join, wrong grain, wrong definition of 'active customer'. A semantic layer is where those definitions already live, and Dot querying through it rather than around it is the reason to prefer this over a generic SQL connector. Zero data retention with the model providers is the other reason it survives a security review.
Dot's endpoint over their data-analysis agent. Two regional addresses, US and EU. It queries your warehouse through Dot's existing semantic layer and permission model rather than issuing raw SQL.
Business-data questions — sales analysis, customer analysis, period-over-period comparison — answered with all of Dot's permissions applied. Neither the marketing page nor the developer docs enumerate tool names.
OAuth by default, with a browser approval; clients that cannot do OAuth use an X-API-KEY header instead. EU accounts connect to the EU endpoint. Dot documents Zero Data Retention with its LLM providers — prompts are neither stored nor used for training.
Paste a URL, then authorize — add the endpoint to your client, then approve the OAuth consent screen