@tiens.nguyen/gonext-local-worker 1.0.78 → 1.0.80

package/gonext_agent_chat.py

@@ -204,18 +204,27 @@ def _summarize_result(task_text: str, agent_output: str,
         return agent_output
 
 
-def _plain_reply(task_text: str, base_url: str, api_key: str, model_id: str) -> str:
-    """Plain chat completion without any tools."""
+def _plain_reply(messages: list, base_url: str, api_key: str, model_id: str) -> str:
+    """Plain chat completion using the full conversation history."""
+    _THINK_RE_LOCAL = re.compile(r"<think>.*?</think>", re.DOTALL | re.IGNORECASE)
+    chat_messages = [{"role": "system", "content": "You are a helpful assistant."}]
+    for m in messages:
+        role = m.get("role", "")
+        content = m.get("content", "")
+        if role not in ("user", "assistant"):
+            continue
+        if role == "assistant":
+            content = _THINK_RE_LOCAL.sub("", content).strip()
+            if not content:
+                continue
+        chat_messages.append({"role": role, "content": content})
     try:
         from openai import OpenAI
         client = OpenAI(base_url=base_url, api_key=api_key or "local",
                         max_retries=0, timeout=60)
         resp = client.chat.completions.create(
             model=model_id,
-            messages=[
-                {"role": "system", "content": "You are a helpful assistant."},
-                {"role": "user", "content": task_text},
-            ],
+            messages=chat_messages,
             temperature=0.7,
             max_tokens=512,
         )
@@ -239,16 +248,28 @@ 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.
@@ -256,7 +277,7 @@ def run_agent_chat(cfg):
 
     if not needs_agent:
         _log("router: plain chat (no HTTP needed)")
-        answer = _plain_reply(task_text, agent_base_url, agent_api_key, agent_model_id)
+        answer = _plain_reply(messages, agent_base_url, agent_api_key, agent_model_id)
         _log(f"plain reply: {len(answer)} chars")
         _emit({"type": "final", "text": answer})
         return

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.78",
+  "version": "1.0.80",
   "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",