How DewPoint works

Connect approved reporting views once. Get fresh numbers without repeat exports, and compare teams or time periods using the same metric definition.

Does the AI read my entire database?

What the AI sees
Connecting a database doesn’t upload it to the model. During setup and edits, the model can receive column names, statistics, and actual sample values. Agent-assisted checks can also share query results.
When you view a chart
Your database runs the query and calculates the result from matching rows. Opening or filtering a saved chart makes no AI call.
How you control access
Grant a read-only account access to approved tables or views. DewPoint can discover other objects the account can access, even when they are outside the workspace.

Compare workflows

Workflow What you need How you work Output
DewPoint Access to DewPoint, a read-only database connection, and a question Ask DewPoint to draft metrics and charts Dashboard that queries live data
Metabase query builder Metabase Cloud or a self-hosted instance connected to your database Choose joins, filters, and summaries in the query builder Saved question or dashboard
ChatGPT file analysis Uploaded file and question Ask ChatGPT to analyze the file Answer, table, or chart in chat

Two completed East orders add up to $360. See the rows and SQL.

Storage and privacy settings

Your database can stay on a private network as long as the DewPoint server can reach it. Connected tables stay in place. DewPoint stores dashboards, encrypted connection settings, and profile metadata in its Postgres instance. It imports uploaded files into that instance. Use synthetic or redacted inputs if your provider cannot receive real values.

INDEX_AGENT_TABLES=0 disables model-assisted investigation during indexing. Other model-assisted steps can still use sampled values. LLM_LOG_PROMPTS=off and LLM_CONTEXT_LOG=false disable local prompt and context logging. Model requests continue.