Skip to main content
All AI features in mogenius — Agents and Chat — run on the AI models you connect to your cluster. You can register several models and choose which one is the default and which are available in chat. Models are managed centrally under Cluster Settings → AI and shared across the whole cluster.

​
Connecting an AI model

Open Cluster Settings → AI and add a model. Each model has the following fields:
  • SDK — The provider: OpenAI, Anthropic, or Ollama. Choose OpenAI for any OpenAI‑compatible endpoint.
  • Model — The model identifier. Pick from the list the provider returns; if none appear, check the endpoint and authentication.
  • API URL — The endpoint URL. Required for Ollama and for OpenAI‑compatible providers; the standard OpenAI and Anthropic endpoints are used by default otherwise. Any OpenAI‑compatible API endpoint works.
  • API Key — Authentication for the endpoint. Enter a key as plain text and mogenius stores it as a managed Kubernetes secret, or reference an existing secret. Ollama endpoints that require no authentication can be left without a key.
After saving, mogenius validates the connection and the model becomes available to your cluster.
You can add multiple models — for example a fast model for chat and a more capable model for agents. Each is stored as its own resource and can be edited, tested, or removed independently.

​
Default and chat‑enabled models

  • Default — Mark one model as the cluster default. Agents that don’t specify their own model, and new chat sessions, use it automatically.
  • Chat enabled — Toggle whether a model appears in the chat model picker. This lets you keep some models for agents only.

​
Token and budget limits

Limits are configured per model, so each model carries its own budget:
  • Daily token limit — The maximum tokens the model may consume per day across agents and chat (0 = unlimited). Once reached, further work is queued until the limit resets.
  • Token budget per run — A per‑run cap on tokens (0 = unlimited). Agents can override this in their own budget settings.
  • Max tool calls per run — A per‑run cap on how many tool calls the model may make. Agents can override this too.
Each model shows its current usage, and you can reset the usage counter manually at any time.

​
Testing and maintenance

The model list provides administrative controls per model:
  • Test — Runs a connectivity probe against the endpoint and reports whether the model is reachable and correctly authenticated. A readiness badge reflects the current state.
  • Reset usage — Resets the recorded token usage for that model to zero.

​
Next steps