Management — Model Catalog

Navigate to: Dashboard → Model Catalog

What it shows: The complete list of LLM models available in AgentWatch — their names, provider associations, context window sizes, and per-token pricing.

Step-by-step: Auto-discover models from a provider

1. Click Model Catalog in the sidebar.
2. Click Discover Models.
3. Select a provider (e.g., OpenAI).
4. Click Probe — AgentWatch calls the provider’s model list API and imports any models not already in the catalog.
5. Review the newly added models and click Enable on any you want available to employees.

Step-by-step: Add a model manually

1. Click Add Model.
2. Fill in:
   • Model Name — display name (e.g., GPT-4o)
   • External Model ID — the exact string sent to the provider (e.g., gpt-4o)
   • Provider — select from configured providers
   • Context Length — max tokens (e.g., 128000)
   • Input Price — cost per 1M input tokens in USD
   • Output Price — cost per 1M output tokens in USD
3. Click Save.

Step-by-step: Disable a model org-wide

1. Find the model in the list.
2. Toggle Enabled off.
3. Any request specifying this model returns {"error": "model_not_available"}.

Pricing column: Pricing affects cost calculations in Usage Insights and Budget Management. If a provider changes their prices, update the model record here — historical costs are not retroactively recalculated.

How to test:

1. Disable the model you use for testing.
2. Send POST /v1/chat/completions with "model": "<that model>" — should get a 400 model_not_available response.
3. Re-enable the model and resend — should succeed.

API endpoints:

Method	Path	Purpose
GET	/admin/models/catalog	List catalog
POST	/admin/models	Add model
PUT	/admin/models/:id	Update pricing, status
DELETE	/admin/models/:id	Remove model
POST	/admin/providers/:id/probe-models	Auto-discover

---