Model Config
Adjust per-chat model parameters like temperature, max tokens, and sampling settings.
The Model Config panel lets you fine-tune how the active model behaves for the current chat. You open it by clicking the config button next to the model selector in the composer.
Changes made here are per-chat - they do not affect other conversations or global defaults.
Available Parameters
Temperature
Controls how random or deterministic the model's output is. Lower values (closer to 0) make responses more focused and predictable. Higher values make them more varied and creative. A value around 0.7 is a common starting point for general conversation; lower values like 0.2 suit factual or code tasks.
Top P
An alternative sampling strategy to temperature, sometimes called nucleus sampling. The model considers only the smallest set of tokens whose cumulative probability adds up to Top P. Setting it to 0.9 means the model samples from the top 90% of likely tokens. Most providers recommend adjusting either temperature or Top P, not both at once.
Top K
Limits sampling to the top K most likely tokens at each step. Useful on providers and models that support it. A value of 40 is a common default.
Frequency Penalty
Discourages the model from repeating the same tokens that have already appeared in the output. A small positive value (e.g., 0.1–0.5) reduces repetitive phrasing. Set to 0 to disable.
Presence Penalty
Encourages the model to introduce new topics by penalizing tokens that have already appeared at all, regardless of frequency. Useful when you want more varied, exploratory answers.
Max Steps
The maximum number of tool-use steps the model is allowed to take before the response ends. This is most relevant for extension-heavy workflows where the model loops through multiple tool calls. Raising it allows longer agentic runs; lowering it keeps things tighter.
Max Tokens
Sets a hard cap on the number of tokens the model generates in a single response. Leave it empty to use the model's default. Set it explicitly when you want to prevent very long responses or stay within a cost budget.
Seed
A fixed seed makes generation deterministic - the same prompt with the same seed should produce the same output. Useful for debugging or reproducibility testing. Not all providers honor this parameter.
The Indicator Dot
When any parameter has been changed from its default, a small indicator dot appears on the config button. This is a visual reminder that the current chat has non-default settings.
To reset everything back to defaults, open the panel and click the reset button.
Leaving Fields Empty
Leaving a field blank tells AgentOne to defer to the model or provider default for that parameter. This is the recommended approach unless you have a specific reason to change a value - provider-side defaults are usually tuned well for each model.
Related
- Model Selector - switch the active model for this chat.
- Model Metadata - understand context window, output limits, and capability flags.
- Chat Settings - global chat behavior like streaming and submit key.