Model Configuration
N.E.K.O. resolves models by role, not through one global model string. The selected profile supplies defaults; supported values in core_config.json can override individual roles.
| Role | Runtime field |
|---|---|
| Core | CORE_MODEL |
| Conversation | CONVERSATION_MODEL |
| Summary | SUMMARY_MODEL |
| Correction | CORRECTION_MODEL |
| Emotion | EMOTION_MODEL |
| Vision | VISION_MODEL |
| Agent | AGENT_MODEL |
| Realtime | REALTIME_MODEL |
| TTS | TTS_MODEL |
Feature code may derive additional role values from these settings.
Recommended flow
- Select Core and Assist providers in the Web UI.
- Enter credentials for the selected providers.
- Run the UI connectivity checks.
- Only then configure a supported per-role model, URL, or key.
A saved resolved provider URL is reused only while it remains in the current profile's candidate set.
Avoid catalog snapshots
Model IDs, endpoints, thinking controls, token limits, and voice catalogs are provider-specific and change over time. Use the Web UI and config/api_providers.json from the same revision as the running app. Examples are not compatibility promises.
Adding a role or field requires synchronized loader, config-manager, router/UI, tests, and all eight locale files. Do not silently pass unsupported parameters; model wrappers deliberately omit options incompatible with reasoning/extended-thinking APIs.