Status
Use Status to check whether the model configuration RAUM needs is available and healthy for the organization. This is the first place to check when Assistant generation or semantic retrieval suddenly stops working after a provider-key, endpoint, or billing change.Models
Use Models to manage the organization’s configured AI provider/model rows. RAUM separates model purposes because generation and semantic retrieval are different workloads.Generation / all-purpose models
These models perform support reasoning and response generation. Depending on the execution path, RAUM can use the configured model stack for stages such as understanding the request, planning, evaluating evidence, and writing the final answer.Embedding models
Embedding models convert procedures, guidelines, articles, documents, and other searchable content into vectors used for semantic retrieval. A missing or failing embedding configuration can make active content effectively invisible to semantic search even though the content still appears in the dashboard.Multiple model configurations
RAUM can store multiple prioritized model configurations for an organization and distinguish their intended purposes. This allows deployments to use different providers or models for generation and embeddings instead of forcing one provider to do everything. The onboarding wizard already supports this by offering a secondary provider when the primary provider cannot supply the required embedding capability.Bring your own provider
RAUM calls the provider using credentials configured for your organization. The available model list and provider capabilities therefore depend on the provider account you connect. Do not use a static documentation page as the authoritative model catalog. Provider model names, access, deprecations, and regional availability change too frequently for that to be sensible.Use the current model catalog and verification controls shown in your RAUM dashboard for your organization.
Changing a model safely
Before switching a production Assistant workflow to a new model:- verify the provider connection;
- confirm that the required generation or embedding purpose is covered;
- run representative tickets in Copilot Mode;
- use Simulation for tool-backed workflows;
- compare answer quality, latency, and cost in the Assistant dashboard; and
- monitor low-confidence tickets and the Coaching queue after the change.
Changing the embedding model
Embedding changes deserve extra care because searchable knowledge was indexed using an embedding configuration. After an embedding-provider/model change, review retrieval behavior and use the embedding repair/backfill controls exposed by the Sources workspace when required. If an embedding repair fails, RAUM surfaces the provider error rather than pretending that a missing vector is a successfully indexed document.Provider failures and fallback
The model layer can use configured fallback behavior where supported. A fallback is useful for provider availability, but it should not hide a configuration mistake such as selecting an embedding-only model for a generation path. When troubleshooting, distinguish:- invalid/expired credentials;
- provider billing or quota failures;
- unavailable model IDs;
- invalid base URLs;
- a missing model purpose; and
- transient provider availability.
Connect an AI provider
Set up provider credentials and choose generation and embedding models.
Assistant dashboard
Compare cost, quality, autonomy, latency, and usage after model changes.