Change the setting
- Open the model’s settings from Stored.
- Find Preserve Thinking in the model’s behavior settings.
- Turn it on or off.
- Choose Save to keep the preference, or Load to apply the settings while loading the model.
What the setting changes
With Preserve Thinking enabled, a compatible model can receive reasoning from earlier assistant turns. This may help with follow-up questions that depend on an earlier derivation or plan.
Keeping that material also uses more of the context window and can increase prompt-processing work. The effect depends on the model and its chat template; it does not guarantee better answers.
With the setting disabled, compatible templates omit earlier reasoning from later prompts. This does not delete the conversation.
Reasoning versus Preserve Thinking
| Control | Purpose |
|---|---|
| Reasoning | Controls whether a supported model reasons before answering the current request. |
| Preserve Thinking | Controls whether earlier reasoning is retained in subsequent conversation context. |
Changing one control does not automatically change the other.
Compatibility
Reasoning-history behavior is available for supported GGUF models whose chat templates honor it. Other model formats and unsupported templates may not provide the same behavior.