Audit & Traces

Governed analytics has to be answerable: who asked what, what the platform did, and whether you can get the same answer again. Colrows records API activity, keeps AI conversation history, shows the reasoning behind every answer, and lets a governed result be re-fetched by ID.

Answer provenance

Every AI answer carries its own explanation - this is what makes Colrows results point-in-time reproducible. Alongside the result of an MCP askData call you get:

• the generated SQL;
• a confidence score and an explanation;
• the semantic sources used - entity types with their names and descriptions (internal IDs are omitted);
• the columns, rows, row count, and a truncated flag;
• the execution ID, when the query ran as a governed execution.

In the AI Data Analyst, each answer's Sources panel lists the metrics, terms, tables, verified queries, and SQL resolution cache entries the answer used, each with its label and description. Each source reference records the source type, its ID, and the source version the answer was built from where one applies, and it is stored with the answer in conversation history.

Conversation & query history

AI conversations are stored per user and can be listed and reopened - the full thread, each interaction's rendered widgets, and the ability to delete a conversation. A governed query execution can be re-fetched by its execution ID, so a result can be reused or re-displayed exactly as it was, rather than re-derived.

API activity audit

Every API call except CORS preflight (OPTIONS) requests is recorded server-side after its response - the request URL, HTTP method, response code, request body, organization, user, and timestamp. Together with conversation history, this is the audit trail behind a governed deployment. MCP tool calls are recorded through privacy-safe telemetry that never logs tokens, natural-language questions, generated SQL, or result rows.