Restore dynamic system prompt via RunnableLambda prompt parameter

Uses prompt=_build_prompt (a RunnableLambda) so LangGraph passes the
runtime config through, allowing per-request system_prompt injection.
Restores user memory loading for web and iMessage plain-text enforcement.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

blueprints/conversation/__init__.py

@@ -26,6 +26,7 @@ from .logic import (
     persist_agent_tool_messages,
     tool_result_text,
 )
+from .memory import build_system_prompt_with_memories
 from .models import (
     Conversation,
     PydConversation,
@@ -142,12 +143,14 @@ async def query():
     )
 
     with time_response(stream=False):
+        system_prompt = await build_system_prompt_with_memories(str(user.id))
         payload = {"messages": [_build_user_message(query)]}
         agent_config = {
             "configurable": {
                 "user_id": str(user.id),
                 "is_admin": user.is_admin(),
                 "thread_id": str(conversation.id),
+                "system_prompt": system_prompt,
             }
         }
 
@@ -234,12 +237,14 @@ async def stream_query():
             logging.error(f"Failed to load image {image_key}: {e}")
             query_text = f"{query_text or ''}\n\n[An image was sent but could not be loaded.]".strip()
 
+    system_prompt = await build_system_prompt_with_memories(str(user.id))
     payload = {"messages": [_build_user_message(query_text or "", image_block)]}
     agent_config = {
         "configurable": {
             "user_id": str(user.id),
             "is_admin": user.is_admin(),
             "thread_id": str(conversation.id),
+            "system_prompt": system_prompt,
         }
     }
 

blueprints/conversation/agents.py

@@ -6,7 +6,8 @@ from dotenv import load_dotenv
 from langchain.chat_models import BaseChatModel
 from langchain.tools import tool
 from langgraph.prebuilt import create_react_agent
-from langchain_core.runnables import RunnableConfig
+from langchain_core.messages import SystemMessage
+from langchain_core.runnables import RunnableConfig, RunnableLambda
 from langchain_openai import ChatOpenAI
 from tavily import AsyncTavilyClient
 
@@ -1230,11 +1231,21 @@ if google_gmail_enabled:
     tools.extend([gmail_search_emails, gmail_get_email])
 
 
+@RunnableLambda
+def _build_prompt(messages: list, config: RunnableConfig = None) -> list:
+    """Inject the per-request system prompt, replacing any prior system messages."""
+    system_prompt = (
+        (config or {}).get("configurable", {}).get("system_prompt", SIMBA_SYSTEM_PROMPT)
+    )
+    non_system = [m for m in messages if not isinstance(m, SystemMessage)]
+    return [SystemMessage(content=system_prompt)] + non_system
+
+
 def build_agent(checkpointer=None):
     """Compile and return the LangGraph agent."""
     return create_react_agent(
         model=model_with_fallback,
         tools=tools,
-        prompt=SIMBA_SYSTEM_PROMPT,
+        prompt=_build_prompt,
         checkpointer=checkpointer,
     )

blueprints/imessage/__init__.py

@@ -17,6 +17,7 @@ from blueprints.conversation.logic import (
 )
 from blueprints.conversation.models import Conversation, ConversationMessage
 from quart import current_app
+from blueprints.conversation.memory import build_system_prompt_with_memories
 from utils.strip_markdown import extract_agent_text, strip_markdown
 
 imessage_blueprint = Blueprint("imessage_api", __name__, url_prefix="/api/imessage")
@@ -343,11 +344,18 @@ async def _process_and_reply(
         else:
             user_content = content
 
+        system_prompt = await build_system_prompt_with_memories(str(user.id))
+        system_prompt += (
+            "\n\nIMPORTANT: You are replying via iMessage. Use plain text only. "
+            "Do not use markdown tables, headers, bullet points, bold, italics, "
+            "or any other markdown formatting. Write in natural prose."
+        )
         agent_config = {
             "configurable": {
                 "user_id": str(user.id),
                 "is_admin": user.is_admin(),
                 "thread_id": str(conversation.id),
+                "system_prompt": system_prompt,
             }
         }
         logger.info(f"Invoking LangChain agent for conversation {conversation.id}")