agentram 0.1.29 → 0.1.30

package/agentram/cli.py

@@ -475,9 +475,11 @@ def run_textual_tui(context: McpContext) -> bool:
         def __init__(self, mcp_context: McpContext) -> None:
             super().__init__()
             self.context = mcp_context
-            self.messages: list[str] = []
-            self.busy = False
-            self.suggestions: list[str] = []
+            self.messages: list[str] = []
+            self.busy = False
+            self.prompt_queue: list[str] = []
+            self.processing_task: asyncio.Task[None] | None = None
+            self.suggestions: list[str] = []
             self.setup_done = False
             self.prompt_history: list[str] = []
             self.history_index: int | None = None
@@ -769,23 +771,37 @@ def run_textual_tui(context: McpContext) -> bool:
                 self.chat.clear()
                 self.chat.write("system > Chat cleared.")
                 return
-            if self.busy:
-                self.chat.write("agentram > busy, wait current task")
-                return
-            self.busy = True
-            self.chat.write("agentram > thinking...")
-            try:
-                response = await asyncio.to_thread(handle_chat_input, self.context, text, self.messages)
-            except Exception as error:  # noqa: BLE001 - TUI boundary
-                response = f"error: {error}"
-            finally:
-                self.busy = False
-            if response == "__clear__":
-                self.chat.clear()
-                self.chat.write("system > Chat cleared.")
-            elif response:
-                self.chat.write("agentram > " + response)
-            self.refresh_sidebar()
+            if self.busy:
+                self.prompt_queue.append(text)
+                self.chat.write(f"agentram > queued #{len(self.prompt_queue)}")
+                return
+            self.prompt_queue.append(text)
+            self.processing_task = asyncio.create_task(self.process_prompt_queue())
+
+        async def process_prompt_queue(self) -> None:
+            if self.busy:
+                return
+            self.busy = True
+            try:
+                while self.prompt_queue:
+                    text = self.prompt_queue.pop(0)
+                    pending = len(self.prompt_queue)
+                    suffix = f" ({pending} queued)" if pending else ""
+                    self.chat.write(f"agentram > thinking...{suffix}")
+                    try:
+                        response = await asyncio.to_thread(handle_chat_input, self.context, text, self.messages)
+                    except Exception as error:  # noqa: BLE001 - TUI boundary
+                        response = f"error: {error}"
+                    if response == "__clear__":
+                        self.chat.clear()
+                        self.chat.write("system > Chat cleared.")
+                    elif response:
+                        self.chat.write("agentram > " + response)
+                    self.refresh_sidebar()
+            finally:
+                self.busy = False
+                self.processing_task = None
+                self.query_one("#input", Input).focus()
 
     AgentRAMTextualApp(context).run()
     return True
@@ -1188,7 +1204,7 @@ def supervisor_workflow_text(context: McpContext, prompt: str) -> str:
     main_agent = agents.get("main")
     small_agent = agents.get("small")
     memory_context = retrieve_text(context, prompt, task_id="codex", limit=8)
-    route = classify_agent_runtime_route(context, prompt)
+    route = classify_agent_runtime_route(context, prompt, main_agent)
     intent = str(route.get("intent", "chat"))
     outputs: dict[str, str] = {}
 
@@ -1252,7 +1268,28 @@ def supervisor_workflow_text(context: McpContext, prompt: str) -> str:
         ]
     )
 
-def classify_agent_runtime_route(context: McpContext, prompt: str) -> dict[str, object]:
+def classify_agent_runtime_route(context: McpContext, prompt: str, main_agent: AgentModelEndpoint | None = None) -> dict[str, object]:
+    if main_agent:
+        try:
+            response = call_agent_endpoint(main_agent, build_agent_runtime_router_prompt(prompt))
+            data = json.loads(extract_json_object(response))
+            intent = str(data.get("intent", "")).strip().lower()
+            if intent in {"planning", "execution", "memory", "chat"}:
+                return {
+                    "intent": intent,
+                    "confidence": float(data.get("confidence", 0.75)),
+                    "reason": str(data.get("reason", "main-agent router")),
+                    "source": "main-agent",
+                }
+        except Exception as error:  # noqa: BLE001 - router boundary
+            fallback = heuristic_agent_runtime_route(prompt)
+            fallback["source"] = "heuristic"
+            fallback["router_error"] = str(error)
+            return fallback
+
+    return heuristic_agent_runtime_route(prompt)
+
+def heuristic_agent_runtime_route(prompt: str) -> dict[str, object]:
     lower = prompt.lower()
     execute_words = [
         "bắt đầu làm", "bat dau lam", "làm đi", "lam di", "tiến hành", "tien hanh", "thực hiện", "thuc hien",
@@ -1263,12 +1300,29 @@ def classify_agent_runtime_route(context: McpContext, prompt: str) -> dict[str,
         return {"intent": "execution", "confidence": 0.9, "source": "heuristic"}
     if any(word in lower for word in plan_words):
         return {"intent": "planning", "confidence": 0.85, "source": "heuristic"}
-    decision = classify_prompt_intent(context, prompt)
+    decision = heuristic_intent(prompt)
     intent = str(decision.get("intent", "chat"))
     if intent == "coding":
         decision["intent"] = "execution"
+    decision["source"] = "heuristic"
     return decision
 
+def build_agent_runtime_router_prompt(prompt: str) -> str:
+    return "\n".join(
+        [
+            "You are AgentRAM orchestration router using the main agent model.",
+            "Classify which agent should handle the user prompt. Do not edit files. Do not run tools. Return only JSON.",
+            "Schema: {\"intent\":\"planning|execution|memory|chat\",\"confidence\":0.0,\"reason\":\"short\"}",
+            "planning = user asks to plan, design, set roadmap, discuss goal, no execution yet.",
+            "execution = user approves plan, says start/do it, or asks coding/refactor/fix/test/file work.",
+            "memory = user asks to remember/save/note a decision/fact.",
+            "chat = normal conversation.",
+            "Nếu user prompt bằng tiếng Việt, reason dùng tiếng Việt.",
+            "User prompt:",
+            prompt,
+        ]
+    )
+
 def runtime_execution_plan(context: McpContext, supervisor: AgentModelEndpoint | None, prompt: str, memory_context: str, route: dict[str, object]) -> str:
     state = context.goal_store.load().context_block()
     if state:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentram",
-  "version": "0.1.29",
+  "version": "0.1.30",
   "description": "Async, model-agnostic Working RAM for coding agents.",
   "license": "MIT",
   "homepage": "https://github.com/trancongnghia/AGENT_RAM#readme",