Monitoring
What one gateway lets you see — every user, every query, and what the engine does with it
Trust requires transparency. Because every query from every surface passes through one gateway, Credible is a single place to see who is asking what, how the model is being used, and what it costs — visibility over every user and every query.
The Audit Trail
Every query is logged to an immutable audit trail: the caller (a user's verified identity or an API key's group), the surface it came from (workspace, MCP, data app, REST API), the model and package version it ran against, and the query itself. This is the record compliance asks for, and it is the same record whether the question came from a person or an agent. Audit logging is part of the Enterprise plan.
Usage and Cost
Credible meters in the units your stack already bills in — tokens, compute time, and storage — and the same gateway that enforces access is where those meters are read. Every meter starts with a monthly allowance; see pricing for the rates. Queries that run directly on your warehouse incur no compute charge from Credible, so the compute meter is a direct view of how much work the engine is doing on your behalf.
What the Engine Does With It
The engine watches every query from every consumer and uses what it sees:
- Performance and cost — query cost and latency across all consumption guide materialization and caching. The hot paths are the ones worth keeping warm.
- Model quality — every retrieval is a test of whether the model surfaced the right concepts. Misses point at a missing
#(doc)line or an unindexed dimension; see Discovery Metadata. Automated tuning from these signals is in preview. - Lineage — find the code that defines any metric and trace an answer from the source database to the number on the screen, because there is one definition and one path to it.