MODEL HUB
Every model. Every claim. Audit-ready.
AI assets inside a regulated enterprise are not just models. They are decisions, with provenance, with evaluation history, with policy bindings, with audit trails. Model Hub is the catalog that treats them that way.
Why a registry, and not a list.
A list of models is what you get from a public model garden. A registry is what your auditor wants. The difference is everything.
When a regulator asks why a particular decision was made — a loan denial, a fraud flag, a medical recommendation — they are asking three questions in one. *Which model produced this output? What were that model's known limitations? Who approved it for production?* If your AI infrastructure cannot answer all three for every model in production, you do not have an AI program. You have an exposure.
Model Hub exists to answer those three questions for every asset, every time.
What's inside an asset record.
Every entry in Model Hub carries six categories of metadata, all required, all auditable.
Identity — name, version, checksum, owner, lifecycle state. The model is uniquely identifiable across deployments and time.
Provenance — base model, training data corpus (with approval reference), trainer, training timestamp, full lineage back to ancestor models. The model's history is reconstructable.
Evaluation — accuracy on benchmarks, fairness measurements across protected groups, red-team results, drift alarms, full evaluation history. The model's behavior is measured, not assumed.
Policy bindings — guardrails, data classification, retention requirements, jurisdictional restrictions, human review requirements. The model knows the rules it operates under.
Audit trail — approvals (named, timestamped), deployment events (cryptographically chained), access logs, regulatory framework mappings. The model's path to production is reviewable.
Lifecycle — current state from `DRAFT` through `RETIRED`, with state transitions logged and reversible. The model has a story, not just a status.
A model that does not have all six is not in production at any of our customers.
What lives in commercial Model Hub. What lives in AgentAnywhere Core.
AgentAnywhere Core (open source) gives you:
- The Model Hub data structure — identity, version, checksum, basic provenance
- The registry API — register, fetch, list, version models
- Local file-based storage and a developer-friendly CLI
- Self-hosting on any conformant infrastructure
Commercial Model Hub adds:
- Full governance metadata — fairness measurements, red-team integration, drift alarms, policy bindings, retention rules
- The audit trail — cryptographic chain, approval workflows, regulatory mappings
- RBI FREE-AI compliance mappings, HIPAA documentation, GDPR data-subject responses
- Enterprise SSO, role-based access controls, multi-tenant deployments
- Integration with Guardrails, TrustFabric, Validator, and Observe — the full governance stack
- Production-grade storage backends, multi-region replication, disaster recovery
- 24×7 support, SLAs, and architecture review
This boundary is intentional. The open-source registry is enough to learn the platform, build prototypes, and contribute upstream. The commercial product is what you deploy when a regulator is going to look at your AI inventory.
Model Hub doesn't sit alone.
Every other capability in AgentAnywhere reads from Model Hub.
Flow Studio and Agent Lab select models for agent steps directly from Model Hub. The metadata available at design time — fairness, drift, policy bindings — informs which models are appropriate for which roles. Models inappropriate for a deployment context are filtered out, with the reason cited.
Guardrails enforces policy bindings at runtime. A model registered with `human_review: REQUIRED for adverse outcomes` cannot be used to make adverse decisions without a human-in-the-loop step in the agent flow.
TrustFabric monitors model outputs in production against the evaluation baselines stored in Model Hub. Drift detection compares live behavior to the registered evaluation history.
Observe writes back to Model Hub. Every production interaction with a model becomes part of that model's audit trail and contributes to drift measurement.
The registry is the spine. The platform is the body that hangs from it.
What you stop maintaining.
Most regulated enterprises building AI today have a half-built model registry — a spreadsheet, a wiki, a folder structure, a Confluence space, a homegrown service. Three teams maintain three views; none of them is current. When the auditor asks, the team scrambles.
Model Hub is what you stop maintaining. It is the registry, the audit document, the approval workflow, the drift monitor, and the regulatory mapping in one place — all generated as a byproduct of normal platform use, not as a separate compliance exercise.
Talk to our team.
Commercial Model Hub is sold as part of an AgentAnywhere deployment. Pricing is structured around scale, sovereignty requirements, and support tier.