Models

Which LLMs are in production — and who approved them?

Every model in use across the enterprise, inventoried and governed: provider, version, hosting, data residency, approval status, risk and cost. Approve, block or pin versions centrally.

Somewhere in your company right now, a team is calling an LLM you've never vetted, sending customer data to a region your legal team hasn't cleared, on a bill no one is watching. Multiply that by every team shipping AI, and "which models are we using?" becomes a question nobody can answer — until an incident, an invoice, or an auditor forces it.

The problem: shadow models, silent cost, hidden residency risk

Models are the easiest part of the AI stack to adopt and the hardest to see. A one-line config change swaps GPT for an open model on someone's GPU. A prompt sends regulated data to a US endpoint. A new version ships with different behavior overnight. Each is invisible to Security, Finance and Legal — until it isn't.

Why it matters to the enterprise

Three stakeholders care, and today none of them have answers. Legal & Compliance need to know where data goes and which models are sanctioned (EU AI Act, data-residency rules). Finance needs to see spend per model before it surprises the quarter. Security needs to block unvetted or compromised models fast. A model inventory is the shared source of truth that lets all three say "yes" or "no" with confidence.

"Customer data may only use EU-hosted, approved models" is a great policy — but only if you can see, and enforce, which models your agents actually call.

How AuthSpoke does it

FinOps for AI, built in

Model usage is the fastest-growing line item most enterprises don't track. Seeing cost and agent-count per model turns "why did our AI bill triple?" into a dashboard you check, not a postmortem you write.

What you get

Govern every model your AI touches

Inventory your models, set approvals and residency rules, and put cost and risk on one screen.