Resources
Configured capabilities available to Agents.
A Resource is a named capability that an Agent can use, such as a model, database, external API, browser, sandbox, or storage location. It keeps connection settings, authentication references, available actions, and access rules together. Agents bind to Resources by name; they do not contain connection details or secret values in their code.
Categories
| Category | Typical Resources |
|---|---|
| AI & Models | Logical Models and the connection Gateways through which they run |
| Apps & APIs | SaaS integrations, MCP servers, web access, and browsers |
| Data | Databases and query services |
| Storage | Artifacts, files, and managed secret capabilities |
| Compute | Sandboxes and other execution environments |
Credentials and Resources solve different problems. A Credential proves identity to an external system. A Resource defines what an Agent can do with that system. One Resource may reference a Credential, and the same Credential can support more than one Resource when Policy permits it.
Gateways, Models, and Tools
The Console organizes Resources into three focused views. Gateways are ordinary invokable Resources explicitly configured as connections; they hold provider settings, Credential references, billing ownership, capabilities, and access rules. Models are logical Agent-facing capabilities configured on top of fixed or scoped Gateways. Tools are Agent-callable capabilities deployed with Agents. Agents bind Models and other Resources by stable identity instead of embedding provider infrastructure in their code.
Choose a path
| Goal | Guide |
|---|---|
| Add or review a capability | Configure a Resource |
| Configure model routing, BYO access, or per-tenant pricing | Gateways and Models |
| Make a capability available to an Agent | Use Resources from Agents |
| Set up authentication for an external service | Credentials |
| Understand an Agent's current bindings | Operate Agents |
The Resource CRN is its stable address. Display names make the Console easier to scan, while bindings and Policy use the exact CRN so similarly named capabilities cannot be confused.