@crossworks/voice-client 0.600.2 → 0.607.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@crossworks/voice-client",
3
- "version": "0.600.2",
3
+ "version": "0.607.0",
4
4
  "description": "Browser-safe surface of the voice/model layer — provider catalogue, model catalogs, audio tags, and the adapter type/metadata contract. Zero deps by design: nothing here may reach the network adapters or node builtins (the jackdaw-repo-split P0 boundary).",
5
5
  "type": "module",
6
6
  "exports": {
@@ -214,9 +214,10 @@ export interface ChatResult {
214
214
  * savings on the responder path. */
215
215
  cacheReadTokens?: number;
216
216
  /** Tokens *written* into the provider's prompt cache, billed at the
217
- * cache-write rate (Anthropic charges ~1.25× input for these). Only
218
- * Anthropic surfaces this distinctly (`cache_creation_input_tokens`);
219
- * every other provider folds it into `tokensIn`. */
217
+ * cache-write rate (Anthropic charges ~1.25× input for these). Anthropic
218
+ * (`cache_creation_input_tokens`) and OpenRouter (`cache_write_tokens`,
219
+ * taken out of `tokensIn`) surface this distinctly; every other provider
220
+ * folds it into `tokensIn`. */
220
221
  cacheWriteTokens?: number;
221
222
  /** Provider-reported USD cost for the call, in dollars (not micro-USD).
222
223
  * Currently only OpenRouter populates this via `usage.cost` — and it
@@ -617,6 +618,12 @@ export interface VisionExtractOptions extends ProviderCredential {
617
618
  model?: string;
618
619
  /** Max output tokens. Vision-LLMs can run long without one. */
619
620
  maxTokens?: number;
621
+ /** Ask the model to think as little as it can: a reading task, where a
622
+ * thinking model's hidden thinking tokens would fill `maxTokens` and cut
623
+ * off the answer. Sent where the adapter knows the provider's setting
624
+ * (OpenRouter: `reasoning.effort` `minimal`, measured 2026-11-11 to keep a
625
+ * model with thinking off as it is); other adapters leave it out. */
626
+ leastThinking?: boolean;
620
627
  }
621
628
 
622
629
  export interface VisionExtractResult {