Two different records get asked for in an audit: "what AI do we run and who approved it?" and "what did this specific call do, and can you prove it?" Collibra answers the first — an enterprise data catalog, lineage, and quality platform, extended into AI Governance: a registry of AI use cases, models, and agents with risk tiering, an aggregate trust score, and compliance-assessment templates (EU AI Act, NIST AI RMF). It is not an LLM gateway; nothing in its product pages claims request-path interception, runtime enforcement, or per-request cost tracking. Wardin answers the second: every gateway-routed call passes through it before a provider, so budget hard-stops, policy blocks, and rate limits are real, and the same in-path decision is written into a cost-attributed, ED25519-signed receipt carrying machine-readable framework-control citations — evidence about each call, not documentation about the system. Most enterprise programs plausibly want both.
For the program that needs a system of record for every AI initiative — what exists, what data feeds it, who approved it — Collibra is the mature, purpose-built answer, at a catalog/lineage/quality scale Wardin isn't built for and isn't trying to replace, with a formal use-case registry, risk tiering, and an aggregate trust score. Ours is narrower: an evidence-bound inventory of gateway-routed use cases with registered-vs-observed drift, not an org-wide intake and approval workflow. Where its scope ends is the traffic itself — it documents the system, it was never sitting between the app and the model.
Wardin is in the request path, so it does what a catalog can't: stop a call over budget or policy before the provider sees it, meter its exact cost to a team, key, and session, and sign a tamper-evident receipt of the decision. Each gateway-routed receipt carries machine-readable control citations across six framework packs (EU AI Act, NIST AI RMF, ISO 42001, SOC 2, NIST 800-53, ISO 27001) — DRAFT mappings, counsel-review-pending, scoped to gateway-routed traffic, and evidence for your own audit rather than a certification. Collibra produces documentation and lineage about a system; Wardin produces a signed record of every individual call.
Choose Collibra if you need an enterprise-wide data catalog, lineage, and quality platform and a formal registry to document and risk-assess every AI use case in the organization. Choose Wardin when the traffic itself has to be governed — a call actually stopped over budget or policy, cost attributed to a team and key, and a signed, framework-mapped receipt for every gateway-routed request. Many programs will reasonably run both: Collibra as the record of what AI exists and is approved, Wardin as the record of what each call actually did — and the one that can prove it.