Parameters
The Parameters tab exposes advanced generation settings for the Custom Model. These settings directly influence how the underlying Provider Model generates responses — affecting output style, consistency, length, and behavior.
The default values are calibrated to work well for most analytical use cases. Changing parameters without a specific, well-understood reason is likely to cause unintended side effects — including inconsistent responses, lower output quality, or unpredictable behavior.
This is especially important for Custom Models that are already in production use with charts and dashboards built on top of them. A parameter change affects every query made through the model from that point forward. Existing charts may render different results, and users will see changed behavior without any visible indication that the model was modified.
If in doubt, leave the parameters unchanged.
When Parameters May Be Worth Adjusting
Parameter changes are occasionally justified in specific, deliberate scenarios — for example:
- A domain expert determines that responses are too verbose and wants to constrain output length
- A specific use case requires more deterministic, highly consistent responses and lower randomness is desirable
- A Builder is experimenting with a new model configuration in a development environment before it is shared with users
In all cases: test thoroughly before sharing the model with users, and document the reason for the change.
Available Parameters
The parameters shown depend on the connected Provider Model and its capabilities. Not all parameters are available for every provider or model type.
Common parameters include generation controls such as temperature, sampling settings, and stop sequences. Refer to the documentation of the underlying Provider Model for the specific meaning and valid ranges of each parameter.
Notes
- Parameter changes take effect immediately after saving — there is no staging or preview.
- Changes are recorded in the Audit Log tab.
- Use the Locked flag on the model to prevent further parameter changes once a stable configuration is in production.