Resources
Resources is a read-only inventory of everything in your tenant, across all projects, in one place: agents, models, tools, knowledge modules, guardrails, providers, sandbox templates and more. It answers questions that are hard to answer project by project: what exists, who changed it last, what does this agent depend on, what breaks if I delete this model, and which references are already broken.
You cannot create, edit or delete anything on this page. To change a resource, open it in its own service. Operators find the page under Settings → Resources.
Who can open it
The Resources page is available to workspace owners and admins. Other members do not see it, and a direct visit to the page returns not found. The role is read from the account's current role, so a promotion or demotion applies on the next navigation.
The inventory API behind the page is a resources service. By default only owners and admins reach it; anyone else needs an explicit grant, which is tenant-wide and lists every resource type and project regardless of the grantee's other permissions. See Permissions reference.
What you see on the screen
The page has two tabs: Resources and Health.
Resources tab
- A list on the left groups resource types by category: Agents, Knowledge, Models, Safety, Quality, Operations and Infrastructure, with a count per type and an All resources entry. Types include agents, tools, models, MCP servers, skills, prompts, memory stores, browser profiles and flows, knowledge modules, vector indexes, file buckets, crawlers, rerankers, providers, inference servers, guardrails, word lists, PII policies, evaluation targets, datasets and suites, red-team campaigns and probes, analysis definitions, alert rules, config groups and items, quota policies, sandbox templates and volumes, MCP hubs, GPU hosts and pools, realtime models, AI App Gateways, prompt optimizations, portals and channel deployments. Enterprise-only types carry an Enterprise badge.
- A table shows name, type, project, status, owner and last update. Search by name, key or ID, and filter by Project and Status. The project filter also offers Not bound to one project. Switch on Only with issues to see just the resources that have a reference problem; an issue count appears on the row.
- The Project column tells you the scope of each resource: a named project, All projects (tenant-wide, usable from every project), Tenant-level (not tied to projects), or Not assigned to a project (so it cannot be used).
Select a row to open the Resource details drawer. It shows:
- the type, key, ID and CRN (a resource name you can copy), the project, the status, and who created and last updated it and when;
- Attributes specific to the type;
- Issues found for this resource, if any;
- Uses and Used by: what this resource references and what references it, each with a status such as OK, Missing, Other project, Inactive or Ambiguous. Use this before deleting or renaming something to see the impact;
- an Open button that switches to the resource's project and opens its own page. If the project cannot be switched, the page warns you and offers Open anyway.
The inventory is read from a short-lived snapshot (about 30 seconds), and refreshing is rate limited, so a just-created resource can take a moment to appear. If a resource type holds a very large number of rows, only the most recently updated rows are listed and a notice says which type was capped. If some types could not be loaded, a notice lists them.
Health tab
The Health tab analyses the references between resources and lists the problems it finds, with an error and warning count. Each issue has a severity, a kind, the resource, the field it concerns, and an explanation. You can filter by severity, issue kind and resource type.
| Issue | Meaning |
|---|---|
| Broken reference | The target does not exist. Depending on the resource it is skipped at runtime, or the resource fails or runs without a safety control. |
| Cross-project reference | The target exists in another project and cannot be used because of the project boundary. |
| Inactive reference | The target is disabled and is skipped at runtime. |
| Ambiguous reference | The same key exists on more than one record, so which one is used is undefined. |
| Duplicate key | More than one record in the same scope uses the same key. |
| MCP registered twice | The same MCP server is registered as both a tool and an MCP server. |
At most the first 1000 issues are shown. If the reference analysis itself could not run, the tab says so, and an empty list then does not mean everything is fine.
Common tasks
- Find where a model is used. Search for the model, open it, and read Used by.
- Check before deleting. Open the resource and look at Used by; anything listed there is affected. Delete confirmations in several services show a similar impact summary.
- Clean up a broken agent. Switch on Only with issues, or open the Health tab, filter by Broken reference, and open the resource to fix the missing target.
- Audit one project. Pick the project in the Project filter.
Permissions and requirements
| Topic | Behaviour |
|---|---|
| Access | The page is for owners and admins. The inventory API also accepts an explicit tenant-wide grant on the resources service. |
| Scope | Tenant-wide, across all projects. The tenant always comes from your session. |
| Read only | No create, edit or delete operations. |
| Usage panels elsewhere | The Where it is used panels in other services answer only with what your own projects may see, and follow the permissions of the service that owns the resource. |