Skip to main content

AI data and privacy in Coginiti

Coginiti's AI features work by sending relevant context to whichever AI provider you or your admin has configured, and by storing your conversations so you can come back to them. This page explains exactly what leaves your environment, where it goes, and how long it's kept - so you know what to expect before you ask.

Background​

Every AI feature in Coginiti - the Forge and Guide chat assistants, inline SQL actions, and AI-powered error explanations - needs to send some of your context to a language model to produce an answer. What gets sent depends on the feature you're using, not on a single global setting. This page covers the three questions that matter most for data governance: what reaches your AI provider, what reaches web search, and what Coginiti stores on your behalf.

What's sent to your AI provider​

Whenever you chat with Forge or Guide, use an inline SQL action (edit, explain, or optimize), or ask AI to explain a failed query, Coginiti sends the relevant content to whichever AI provider is configured for your account or organization (see supported providers). Depending on what you're doing, this can include:

  • Your message or instruction - what you typed
  • Your custom prompt - if you've set one under your personal AI Assistant settings, it's included as a standing instruction on every request
  • The SQL or file content you're working with - the query or tab you have open or selected
  • Table, column, and schema details, including data types - when metadata sharing is relevant to your request (see the metadata sharing toggle)
  • Project and catalog entities - items from your Project Hub or catalog, when they're relevant to answering your request
  • Semantic layer schema - entities, dimensions, and measures from a semantic model, when a request needs to run or plan a semantic query
  • The failed query and its error message - when you ask AI to explain an error
  • Query results - when a request runs a query and the AI needs the results to answer accurately
  • Recent conversation history - since a chat is a multi-turn conversation, earlier messages in the same chat are included as context

None of this is filtered or summarized before it's sent - whatever's relevant to your request goes to the provider as you see it. If you need to keep a particular database or schema out of the AI's reach entirely, use the connection and database filter described in Restrict what the AI can see - it's enforced on the server, so an excluded database is unreachable no matter what you ask.

When Forge or Guide uses its web search tool to look something up, only the search terms it generates for your question are sent out - to Coginiti's web search provider - not your data, your account details, or the rest of your conversation. The provider returns a list of results, which the assistant summarizes back into the chat.

Chat storage and retention​

Every chat conversation - your messages, the assistant's responses, and anything you attached as context - is stored in Coginiti's own database, tied to your user account. It isn't stored by your AI provider.

There's currently no automatic expiration: a chat stays available until you delete it yourself (see manage chat history). Deleting your Coginiti account removes your chat history along with it.

Sharing a chat​

Sharing a conversation (Team and Enterprise only) creates a frozen snapshot of the chat as it exists at the moment you click Share - anything you add to the conversation afterward isn't visible to the person you shared it with, unless you share it again. Use Revoke at any time to remove their access immediately; see share a chat for the click-through steps.


tip

For more information about other exciting Coginiti features visit our documentation

Thank you for choosing us!​