@agentionai/agents 1.9.0 → 1.10.0

package/dist/agents/llamacpp/LlamaCppAgent.d.ts

@@ -99,6 +99,18 @@ export declare class LlamaCppAgent extends OpenAICompatibleAgent {
      * included, is on `raw`.
      */
     listModels(): Promise<ModelInfo<LlamaCppModelCard>[]>;
+    /**
+     * Tells the server to end the model's reasoning phase early, mid-stream,
+     * via llama.cpp's proprietary `/chat/completions/control` endpoint. Useful
+     * for cutting off a model that is thinking for too long without waiting
+     * for it to decide to stop on its own.
+     *
+     * A no-op if no streamed completion is in flight yet (`lastChunkId` unset).
+     * Best-effort: a failed request is logged (when `debug` is on) rather than
+     * thrown, since this is a side channel to a turn that should otherwise
+     * proceed normally.
+     */
+    skipReasoning(): Promise<void>;
 }
 export {};
 //# sourceMappingURL=LlamaCppAgent.d.ts.map

package/dist/agents/llamacpp/LlamaCppAgent.js

@@ -68,6 +68,36 @@ class LlamaCppAgent extends OpenAICompatibleAgent_1.OpenAICompatibleAgent {
             },
         }));
     }
+    /**
+     * Tells the server to end the model's reasoning phase early, mid-stream,
+     * via llama.cpp's proprietary `/chat/completions/control` endpoint. Useful
+     * for cutting off a model that is thinking for too long without waiting
+     * for it to decide to stop on its own.
+     *
+     * A no-op if no streamed completion is in flight yet (`lastChunkId` unset).
+     * Best-effort: a failed request is logged (when `debug` is on) rather than
+     * thrown, since this is a side channel to a turn that should otherwise
+     * proceed normally.
+     */
+    async skipReasoning() {
+        if (!this.lastChunkId)
+            return;
+        try {
+            await fetch(`${this.config.baseURL}/chat/completions/control`, {
+                method: "POST",
+                headers: { "Content-Type": "application/json" },
+                body: JSON.stringify({
+                    action: "reasoning_end",
+                    id: this.lastChunkId,
+                }),
+            });
+        }
+        catch (error) {
+            if (this.debug) {
+                console.error(`Failed to signal reasoning_end to ${this.getVendorName()}: ${error instanceof Error ? error.message : "Unknown error"}`);
+            }
+        }
+    }
 }
 exports.LlamaCppAgent = LlamaCppAgent;
 //# sourceMappingURL=LlamaCppAgent.js.map

package/dist/agents/openai-compatible/OpenAICompatibleAgent.d.ts

@@ -35,6 +35,14 @@ export declare abstract class OpenAICompatibleAgent extends BaseAgent {
     protected config: Partial<OpenAICompatibleConfig>;
     private vizEventId?;
     private currentToolCallCount;
+    /**
+     * The `id` of the most recently seen streamed `ChatCompletionChunk`
+     * (`chatcmpl-...`). Internal only — some servers key control-plane calls
+     * (e.g. llama.cpp's `/chat/completions/control`) off the in-flight
+     * completion's id, so it needs to be captured somewhere subclasses can
+     * reach it, but it isn't part of the public `StreamChunk` shape.
+     */
+    protected lastChunkId?: string;
     /**
      * Whether this server accepts a replayed `reasoning_content` field on an
      * assistant message. `undefined` until proven otherwise — see

package/dist/agents/openai-compatible/OpenAICompatibleAgent.js

@@ -401,6 +401,8 @@ class OpenAICompatibleAgent extends BaseAgent_1.BaseAgent {
             // taking the last one covers both layouts without double-counting.
             if (chunk.usage)
                 streamUsage = chunk.usage;
+            if (chunk.id)
+                this.lastChunkId = chunk.id;
             if (chunk.choices.length === 0)
                 continue;
             const choice = chunk.choices[0];

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@agentionai/agents",
   "author": "Laurent Zuijdwijk",
-  "version": "1.9.0",
+  "version": "1.10.0",
   "description": "Agent Library",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",