AI Inventory
The AI Inventory is a governed catalog of LLM models and agent frameworks. AI Stewards use it to decide which models the organization is allowed to use — and any agent that calls an unapproved model is flagged as a compliance violation in the AI Command Center and Risk Dashboard.
Enterprise feature. Browsing is available to governance roles; approving / blocking models requires the Owner or AI Steward role (AI Developers have read-only access). Open it from AI Governance → AI Inventory.
What's in the catalog
- Models — 570+ LLMs across OpenAI, Anthropic, Google, Meta, Mistral, Cohere, and more, each with its provider key (e.g.
openai/gpt-4o), pricing (input/output per 1M tokens), context window, and capabilities. - Frameworks — agent and orchestration frameworks (LangChain, LangGraph, CrewAI, AutoGen, and others) with type, language, license, and documentation links.
Model attributes
| Attribute | Description |
|---|---|
| Manufacturer | OpenAI, Anthropic, Google, Meta, Mistral, Cohere, … |
| Status | current (maintained), deprecated (legacy), or archived (end-of-life) |
| Capabilities | Function calling, reasoning, JSON mode |
| Pricing | Input / output token prices |
| Context window | Maximum context length |
Finding and comparing models
- Search by model name or provider key.
- Filter by manufacturer, status, and capabilities (combine filters freely).
- Compare 2–5 models side by side — pricing, context window, and capabilities — to choose what to approve.
Approving models
Approval is an organization-level decision that defines your allow-list.
Find the model
Search or filter the catalog to the model you want to govern.
Toggle approval
Set the model to Approved (allowed) or leave it unapproved (blocked). Add a governance note to record the rationale — useful for audits.
Review usage
Each approval tracks a usage count and last-used timestamp, so you can see which approved models are actually in use and prune the rest.
Bulk actions. You can approve or unapprove up to 100 models at once with a single governance note — handy for onboarding a provider or retiring deprecated models in one pass.
Why approvals matter
Approvals are the control behind unapproved-model violations: when an agent's traces show it calling a model that isn't on your approved list, that usage is counted and surfaced as a compliance violation on the agent — in the Command Center detail panel and in the agent's risk score. Keeping the inventory curated is what makes those violations meaningful.
Next steps
- AI Command Center — see which agents are violating model policy.
- Risk Dashboard — model risk feeds the overall risk score.
- Compliance — model governance is part of your framework evidence.