01 |WHAT YOU GET
Built to be defended. Each fact, read and answer leaves a record.
Source object
Each fact stores its source as a kind and a trust level, with a reference and span back to the passage or the facts it was derived from, beside its confidence.
Read more
Decision traces
Each write records a decision trace, linked to the facts it touched.
Read more
Retrieval traces
Each ranked read records a retrieval trace, with its per-signal scores.
Read more
Response traces
Each chat and reflect answer records a response trace, linked to the retrievals and facts behind it.
Read more
Audit events
A separate stream, exported through OpenTelemetry, records writes, deletions, forgets, key and configuration changes, background jobs and denied access attempts.
Read more
Retention per tenant
Each tenant sets how long traces are kept. By default decision traces are kept indefinitely and retrieval and response traces for 90 days.
Read more
02 |WHY IT MATTERS
Trust is a storage property. The evidence lives in the tenant's own database.
Queryable
Traces you can question
Traces are linked to the facts they touched and are queryable through the traces endpoint and the inspect MCP tool.
Tenant
Held in your own database
Decision, retrieval and response traces are written to the tenant's own database, beside the facts they describe.
Access
Deny by default
Principals hold per-verb grants and access is denied unless granted. Keys can only narrow a grant, never widen it, and denied attempts appear in the audit stream.
History
Superseded facts stay
A new value closes the validity of the old one instead of overwriting it, so what changed, and when, stays visible.
GET STARTED
Give your agents answers you can defend. Every fact keeps its source, and writes, ranked reads and chat answers are traced in your own database.
