> ## Documentation Index
> Fetch the complete documentation index at: https://docs.flowx.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Registry

> Catalogue every model, deployment, and owner across your portfolio — the inventory layer for governance.

The AI Registry is the inventory layer. Before you can govern a portfolio, you need to know what's in it. The registry catalogues models, where they're deployed, who owns them, and what data they touch.

***

## What the registry tracks

| Entity | Fields |
| - | - |
| **Model** | Name, provider, version, modality, capabilities, intended use. |
| **Deployment** | Environment (dev/stage/prod), endpoint, region, app linkage. |
| **Owner** | A user or team. Multiple owners allowed per model. |
| **Data tags** | Data classifications the model touches (PII, financial, health). |
| **Risk tier** | Manual classification or computed from data tags + capability. |

***

## Why this matters

Governance, compliance, and ROI calculations all need the registry as the inventory of record:

```mermaid theme={"dark"}
flowchart LR
    Registry[AI Registry] --> Risk["Risk Dashboard<br/>(data + model risk)"]
    Registry --> Compliance["Compliance<br/>(scope of high-risk apps)"]
    Registry --> ROI["ROI<br/>(per-model cost analysis)"]
    Registry --> Audit["Audit Trail<br/>(ownership for changes)"]
```

If a model isn't in the registry, it doesn't show up in any of the downstream views. The first job of a new Observatory org is usually populating the registry.

***

## Populating the registry

Three ways:

> [card] From telemetry
>
> Models that show up in runs are auto-registered. You confirm and fill in metadata.

> [card] Manual entry
>
> Add a model that isn't running yet (preview, eval, planned).

> [card] Import
>
> Bulk import from CSV or from FlowX AI Platform's agent definitions.

> [card] API
>
> Programmatic registration through `POST /api/registry/models`.

***

## Risk tiering

The default tiers map roughly to the EU AI Act risk pyramid:

| Tier | Examples |
| - | - |
| **Minimal** | Internal-tool agents, productivity helpers. |
| **Limited** | Customer-facing chatbots without consequential decisions. |
| **High** | Credit decisions, claims approvals, hiring-adjacent. |
| **Unacceptable** | Banned by regulation — flagged for removal. |

Tiering can be manual or computed. Computed tiering uses data tags and capability flags (e.g. "makes decisions about people" + "trained on personal data" → High).

***

## API

| Endpoint | Use |
| - | - |
| `GET /api/registry/models` | List models. |
| `POST /api/registry/models` | Register a model. |
| `PUT /api/registry/models/{id}` | Update metadata. |
| `GET /api/registry/deployments` | List deployments. |
| `POST /api/registry/deployments` | Link a model to an environment/app. |

***

## Related resources

> [card] [Risk Dashboard](https://docs.flowx.ai/6.x/observatory/governance/risk-dashboard.md)
>
> Where data and model tags become risk scores.

> [card] [Compliance](https://docs.flowx.ai/6.x/observatory/compliance/overview.md)
>
> Risk tier drives which frameworks apply.


## Related topics

- [NIST AI RMF](https://docs.flowx.ai/6.x/observatory/compliance/nist-ai-rmf.md)
- [Compliance](https://docs.flowx.ai/6.x/observatory/compliance/overview.md)
- [Risk Dashboard](https://docs.flowx.ai/6.x/observatory/governance/risk-dashboard.md)
- [EU AI Act](https://docs.flowx.ai/6.x/observatory/compliance/eu-ai-act.md)
- [ROI & value](https://docs.flowx.ai/6.x/observatory/roi/overview.md)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.