Audit and evaluation

Überblick

Audit and evaluation* describe how Ottili AI makes AI work traceable and reviewable. Every Ottili AI action — a model call, a mission, a tool call, an approval or a deducted credit — is resolved to exactly one company and recorded through Ottili Core Audit* (core.audit). Audit is live (LIVE) and Core-backed: it is the same platform primitive that Ottili HQ, Ottili Coder, Ottili LD3 and Ottili Cloud also use.

This article explains what is audited in Ottili AI, how the audit trail is produced, how you evaluate AI activity, and what maturity status each capability carries.

What is audited in Ottili AI

Ottili AI runs Ottili's own models — Ottili Vale* and Ottili Cairn* — behind a single OpenAI-compatible API, with adaptive routing, missions, the Dynamic Workforce Engine (DWE) and tool use. Each of these layers produces traceable records:

• Model & routing* — which model (Vale or Cairn) answered a request and how Ottili Auto routed it.
• Missions & DWE* — which goal was framed as a mission and which role-based DWE specialists worked on it.
• Tools* — which tools were called across the Unified API business.* and ld3.* namespaces to act on your data.
• Approvals* — critical, external actions run approval-first; every step of the approval lifecycle is captured.
• Credits & cost* — every request reserves and deducts credits; cost control enforces model and budget limits and rate limits.

All of these events are resolved to exactly one company (company_id). The company is the tenant boundary: your audit data is isolated from that of other companies.

How the audit trail is produced

The foundation is Ottili Core Audit* (core.audit): a shared audit capability emitted over the shared event bus. Ottili AI resolves audit events through the same Core function instead of building its own log. That is what "audit across products" means: no product owns or duplicates the audit model — all reach for the same platform primitive.

Audit records are designed to be append-only: no in-place change, no deletion. We qualify this honestly — it is enforced at the service level (for example, in the audit tables of the participating services), not as a single global guarantee across the entire platform. We communicate the honest guarantee rather than overstate it.

How you evaluate AI activity

"Evaluation" is the process of judging what the AI did. In Ottili AI this does not happen in an isolated dashboard but through the existing, company-scoped records:

1. Read the audit trail* — every action is traceable: who did what, when, with which model and in which company context.

2. Check approvals* — critical, external effects only happened after an approval was granted. The approval lifecycle documents who requested and who approved.

3. Trace cost* — credits and cost control make every request affordable and measurable; you see which workload caused how much effort.

4. Place status* — every capability carries a maturity-status label; you judge availability by the canonical status vocabulary, not by marketing claims.

This turns audit into a basis for evaluating AI behaviour, tracing errors and demonstrating compliance — without any action leaving your company context.

Status and availability

Ottili AI is part of Ottili ONE. The capabilities relevant to audit and evaluation carry a maturity status from the canonical product-status registry (config/product_truth/feature_status.yaml):

The canonical maturity vocabulary covers CONCEPT, PLANNED, IN_DEVELOPMENT, PRIVATE_BETA, BETA, ROLLOUT, LIVE, DEPRECATED and DISABLED. For the audit area of Ottili AI the listed capabilities are LIVE — we do not claim an evaluation capability that is not verified as live.

How Ottili ONE labels maturity publicly is described in [Understand feature-status labels](/docs/understand-feature-status-labels).

Embedded in Ottili ONE

Ottili AI is native to Ottili ONE. Audit shares with the other Ottili surfaces:

• one login* (One Login)
• one company context* (every action belongs to exactly one company, company_id)
• one audit trail* (Ottili Core Audit)

This means an Ottili AI job always runs inside your company; every action is resolved to exactly one company and recorded through Ottili Core Audit*. The same models and routing power Ottili Coder, Ottili LD3 and other surfaces from inside your company. The shared account structure is described in [What is Ottili ONE?](/docs/what-is-ottili-one).

Related articles

• [Dynamic Workforce Engine](/docs/dynamic-workforce-engine)
• [The Ottili Cairn model](/docs/ottili-cairn-model)
• [The Ottili Spire model](/docs/ottili-spire-model)
• [What is Ottili ONE?](/docs/what-is-ottili-one)
• [Understand feature-status labels](/docs/understand-feature-status-labels)