> ## Documentation Index
> Fetch the complete documentation index at: https://docs.gdilabs.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Audit-ready by default

> Cerebrum was designed for places where you need to explain what an AI did and why. Every action is recorded.

## What gets logged

* **Every job.** Status, who submitted it, when, the classification it got, and the agents that touched it.
* **Every escalation.** Why a conflict moved up a level and what attempt it was on.
* **Every paid call.** Model, tokens, cost.
* **Every retrieval.** Which knowledge-hub chunks were pulled, and what citations were produced.

This sits alongside the job status, so the same query that tells you "was this successful?" also tells you "what did it cost?" and "where did the answer come from?".

## Sensitivity tiers

Every document in the knowledge hub carries a sensitivity tag:

* **Public** — anyone can see it (the docs you're reading now).
* **Partner** — visible to a specific partner.
* **Internal** — visible to the team.
* **Secret** — restricted further.

Retrieval respects these tiers. A partner-facing agent only sees what's tagged public or for that partner. An internal agent sees more. Secret-tagged content is excluded from every retrieval unless explicitly allowlisted.

## PII redaction

PII redaction runs at two boundaries: on the way in (before a request hits a model) and on the way out of the knowledge hub (before chunks land in a prompt). The patterns are configurable per organization.

## What this gives you

A defensible trail. When a stakeholder asks "why did the AI do this?", you can answer with the actual classification, the actual sources, the actual decisions — not a reconstruction after the fact.