Check model compatibility

An AI Model is ready to use when its access, lifecycle, transport, capabilities, route, and pricing all line up. A model appearing in a catalog does not mean that every credential, API, or coding harness can use it.

Start in the Developer Portal

Open AI Models and inspect the model details before configuring a client. The portal can show:

• access status: Available, Needs access, Denied, or Unavailable;
• lifecycle status such as Available, Preview, Deprecated, Disabled, or Archived;
• public model ID, context window, maximum output, modalities, and capabilities;
• supported public transports: OpenAI Responses, OpenAI Chat Completions, Anthropic Messages, or Embeddings where applicable;
• organization recommendations for use cases and coding harnesses;
• compatibility status and feature tier for Codex, Claude Code, and OpenCode; and
• organization notes, advisories, and customer-facing rate information when available.

The same credential-scoped model list is available through GET /v1/models. Use the exact returned id in a request. When supported on your platform, use GET /v1/models/{model_id} to inspect one accessible model.

When the portal shows Needs access, use its access-request action when available or ask an organization administrator to review the model-access policy. Denied means the current effective policy does not authorize the model. Unavailable means the model or route cannot currently serve the requested context.

Interpret harness compatibility

Status	Meaning
Supported	The transport and displayed feature tier are compatible, and the certified harness version passed Gateway conformance.
Limited	The transport is compatible, but the pinned harness version has not completed qualification, or the complete agent feature set is not supported.
Unavailable	The model is not accessible, is not declared for the transport, lacks conversational capability, or has no qualified route profile.

The feature tier describes the level of support backed by qualification evidence:

• Text: text requests are supported.
• Agent: text, streaming, and client tool transport are supported by the qualified profile.
• Reasoning: the agent profile also exposes portable reasoning continuation.

Treat Limited as a compatibility signal, not a promise that every client feature will work. Read the harness-specific guide and the model’s organization advisory before using it in production.

Public model IDs and auto

Conscia selects provider-neutral public model IDs. Do not construct them from provider names, provider-native IDs, aliases, slash-qualified names, or internal route IDs.

Use an explicit public model ID when a workload requires a specific model. Use model: "auto" only on a supported transport when the organization has configured automatic selection. auto selects one eligible public model; it does not bypass model access, capability checks, allowances, route readiness, or pricing requirements.

When a model is visible but unusable

Check the following in order:

1. The credential can see the model in GET /v1/models.
2. The model’s access status is Available.
3. The requested API transport appears in the model detail.
4. The requested capability, such as chat, tools, reasoning, or embeddings, is enabled.
5. The harness compatibility status is not Unavailable.
6. The route is operational and has complete customer pricing.

An organization administrator can review effective model access and route readiness. Platform administrators handle provider-native route configuration; callers use the public model ID and transport reported for their credential.