retainsReasoningHistory
function retainsReasoningHistory(kwargs): boolean;Defined in: providers/reasoning-kwargs.ts:365
Whether a resolved chat_template_kwargs set asks the model to retain its
FULL reasoning history across turns — "Preserved Thinking" rather than plain
interleaved thinking. True when a family's continuity key is present: GLM's
clear_thinking: false or Kimi/Qwen's preserve_thinking: true.
The distinction drives how much reasoning the loop resends. Interleaved thinking needs reasoning only on tool-call turns; preserved thinking needs it on EVERY assistant turn, so the resent history matches what the kwarg promises (GLM/Kimi/Qwen want the complete, in-order reasoning sequence back). The loop reads this — via ModelClient.retainsReasoning — to decide whether to keep reasoning on plain (non-tool-call) turns too.
Parameters
| Parameter | Type | Description |
|---|---|---|
kwargs | Record<string, unknown> | undefined | The resolved chat_template_kwargs, or undefined. |
Returns
boolean
true when a preserve-everything continuity key is set.