@tiens.nguyen/gonext-local-worker 1.0.78 → 1.0.79

package/gonext_agent_chat.py

@@ -239,16 +239,28 @@ def run_agent_chat(cfg):
 
     _log(f"start model={agent_model_id!r} base={agent_base_url!r} maxSteps={max_steps}")
 
-    # Use only the latest user message as the agent task.
+    # Build task from the conversation history.
+    # Include prior USER messages as context so the agent has conversational memory,
+    # but exclude prior ASSISTANT messages (they contain raw HTTP/thinking content
+    # that confuses small models).
     task_text = ""
+    prior_user_msgs = []
     for m in messages:
-        if m.get("role") == "user":
-            task_text = m.get("content", "")
+        role = m.get("role", "")
+        content = m.get("content", "")
+        if role == "user":
+            if task_text:
+                prior_user_msgs.append(task_text)
+            task_text = content
 
     if not task_text:
         _emit({"type": "final", "text": "[No user message found in history]"})
         return
 
+    if prior_user_msgs:
+        context = "\n".join(f"- {t[:300]}" for t in prior_user_msgs[-4:])
+        task_text = f"Conversation context (previous user messages):\n{context}\n\nCurrent task: {task_text}"
+
     _log(f"task={task_text[:120]!r}")
 
     # Route: ask the model if this task needs HTTP tool use.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.78",
+  "version": "1.0.79",
   "description": "Polls GoNext cloud API for async local LLM jobs and runs them against Ollama/OpenAI-compatible servers on this Mac",
   "type": "module",
   "license": "MIT",