Choose and enable models

Use Cloudflare’s AI catalog to add a model without waiting for a Crewhelm release. Crewhelm starts with a small catalog; you decide which exact model IDs are enabled for your Agents.

Before you begin

• Full control access.
• A model available in the Cloudflare AI model catalog.
• Any Cloudflare billing or account prerequisites for that model. Third-party models use unified billing; Crewhelm does not infer billing readiness.

browse_models returns only text-generation models that Crewhelm can route with tools. Workers AI entries are live; third-party entries come from an hourly, last-known-good Cloudflare snapshot. Snapshot provenance is returned with every page. OpenAI models use Responses, Anthropic models use Anthropic Messages, and compatible providers use Chat Completions.

Cloudflare descriptions, capabilities, and pricing are untrusted provider data. Enabling a model does not certify its quality.

Discover and enable a model

1. Call crewhelm_inspect_models and select browse_models.
2. Filter by query, provider, platform, task, or capability. Results use relevance order and omit descriptions by default; set includeDescriptions: true for a narrowed comparison. Reuse the same filters with nextPage when another page exists.
3. Call inspect_model with the exact returned model ID. Confirm that availability is available and runtimeCompatibility is compatible. Review its compatibility evidence, provider source, platform, and pricing source.
4. Call list_enabled_models and retain its model-catalog revision.
5. Call crewhelm_change_models with operation.kind: "add_model", the exact modelId, the current expectedRevision, one stable requestKey, and confirm: false.
6. Repeat the unchanged operation with the same requestKey and confirm: true to enable it.
7. Create or replace an Agent with the inference.workers-ai capability and the exact enabled ID as primaryModel. Agent revisions pin that exact ID.

Verify the result

• list_enabled_models includes the exact ID and a newer catalog revision.
• Agent creation returns the exact enabled ID. Execution limits remain independent of model choice.

Remove or change a model

1. Call list_enabled_models and retain the current revision.
2. Call crewhelm_change_models with operation.kind: "remove_model", the exact ID, expectedRevision, one stable requestKey, and confirm: false.
3. Review impact.affectedAgents. This bounded list contains current Agents configured with the model as their primary or a fallback; affectedAgentsTotal reports the complete count.
4. If the model is the current default, include an already-enabled replacementDefaultModelId.
5. Repeat the unchanged operation with the same requestKey and confirm: true.

Removal prevents new Agent admission and Runs that depend on the disabled model. It does not rewrite or delete immutable Agent revisions.

Recover safely

• On revision_conflict, list enabled models again and decide against the new current revision.
• On model_unavailable, repeat exact inspection. Do not substitute a similarly named ID.
• Resolve billing failures in Cloudflare before retrying the same immutable Agent revision.
• If a removal preview is truncated, inspect Agents by exact model ID before confirming.

Next step

Create your first Agent with the enabled model, or review owner responsibilities before enabling a paid model.