Capability resolver

The Capability resolver screen lets you see how a reasoning effort level and the parameters of a call are translated into the native parameters of a model, or why they are rejected, without calling the model. Portal path: /llm/resolver. It is used by developers and AI platform engineers who prepare a call, and by support engineers who explain a refusal. The screen is built in portal/src/screens/llm-tools.tsx (Resolver, resolverCfg, ResolveResult).

What the screen shows

One panel Resolve parameters ("POST /v1/models/resolve with an effort level and call parameters.") in two parts.

The form, on two columns:

Field	Content
Model	Registry id of the model (required).
Effort	`none`, `low`, `medium`, `high` or `max` (required).
Temperature	Number.
Top P	Number.
Max tokens	Number, checked against the model limit.
Response format	`text`, `json_object` or `json_schema`.
Tool choice	`auto`, `none` or `required`.
Stream	Yes or no.
Extra parameters	JSON object merged into the request body (default `{}`).

The Result, after Run:

a banner Accepted (green) or Rejected (red), with the reason when the API gives one;
the warnings of the resolver, one per line;
Native parameters: the JSON sent to the provider for this model;
Raw response, folded, with the full answer.

Before the first run the result area says "Run the request to see the result here."

Actions

Button or action	What it does	API call	Roles
Run	Sends the model, the effort and the parameters (the Extra parameters are merged at the top level) and shows the resolution.	`POST /v1/models/resolve`	LLM readers
Raw response	Unfolds the full JSON answer.	none	everyone

Roles

POST /v1/models/resolve requires one of the READ roles of services/control-plane/src/eai/domains/llm/__init__.py, which are the llmReaders of portal/src/lib/kit.ts: tenant_admin, platform_admin, security_officer, finops, developer, auditor, viewer, agent, ai_platform_engineer, ai_governance_officer, support_engineer.
The form is not restricted by the portal: the server decides.

Typical tasks

Check that a model accepts a reasoning effort

Choose the model. Type its registry id in Model, as shown in the Models registry. Expected result: the field is filled. If it fails: "Model <model_id> not found." means the id is wrong.
Set the effort. Choose Effort high and leave the other fields empty, so that you test the effort alone. Expected result: the form is valid. If it fails: Required under Effort.
Run. Click Run. Expected result: Accepted and the Native parameters the provider receives. If it fails: "Reasoning effort <requested> is not supported by this model." (effort_not_supported); choose a level listed in the Reasoning modes of the model.

The same check from the command line (see docs/sdks/cli-reference.md):

enterprise-ai models --effort high resolve <model-id>

<model-id>: registry id of the model.

Validate the parameters of a structured output call

Fill the call. Set Model, Effort, Max tokens and Response format json_schema, and put the schema in Extra parameters, so that the resolver sees the same body as the real call. Expected result: valid JSON in Extra parameters. If it fails: Invalid JSON.
Run. Click Run. Expected result: Accepted, possibly with warnings. If it fails: "max_tokens is above the model limit of <n>." or "Parameter response_format is invalid." (schema missing); fix the value and run again.

Troubleshooting

Symptom	Cause	Fix
"Model `<model_id>` not found." (`model_not_found`)	Unknown registry id.	Copy the id from the Models registry.
"Reasoning effort `<requested>` is not supported by this model." (`effort_not_supported`, 422)	The level is outside the reasoning modes of the model.	Choose a supported level, or `none`.
"Parameter `<parameter>` is not supported by this model." (`parameter_not_supported`, 422)	The model does not take this parameter.	Remove it from the form or from Extra parameters.
"Parameter `<parameter>` is out of the allowed range." (`parameter_out_of_range`, 422)	Value outside the bounds of the model, for example Temperature.	Use a value inside the range.
"max_tokens is above the model limit of `<n>`." (`max_tokens_exceeded`, 422)	Max tokens above Max output tokens of the model.	Lower Max tokens.
"Parameter `<parameter>` is invalid." (`parameter_invalid`, 422)	Wrong value, for example `json_schema` without schema.	Fix the value named in the message.
"Input modality `<modality>` is not supported by this model." (`modality_not_supported`, 422)	Extra parameters carry an image, audio or video input the model does not take.	Pick a model with this modality.
Invalid JSON under Extra parameters	The field is not a JSON object.	Type a JSON object, `{}` when empty.

What the screen shows​

Actions​

Roles​

Typical tasks​

Check that a model accepts a reasoning effort​

Validate the parameters of a structured output call​

Troubleshooting​

Related pages​