feat: integrate mempalace archival into background review

model_tools.py

@@ -158,6 +158,7 @@ def _discover_tools():
         "tools.send_message_tool",
         # "tools.honcho_tools",  # Removed — Honcho is now a memory provider plugin
         "tools.homeassistant_tool",
+        "tools.mempalace_tool",
     ]
     import importlib
     for mod_name in _modules:

run_agent.py

@@ -1027,6 +1027,7 @@ class AIAgent:
         self._memory_nudge_interval = 10
         self._memory_flush_min_turns = 6
         self._turns_since_memory = 0
+        self._turns_since_palace = 0
         self._iters_since_skill = 0
         if not skip_memory:
             try:
@@ -1035,6 +1036,7 @@ class AIAgent:
                 self._user_profile_enabled = mem_config.get("user_profile_enabled", False)
                 self._memory_nudge_interval = int(mem_config.get("nudge_interval", 10))
                 self._memory_flush_min_turns = int(mem_config.get("flush_min_turns", 6))
+                self._palace_nudge_interval = int(mem_config.get("palace_nudge_interval", 15))
                 if self._memory_enabled or self._user_profile_enabled:
                     from tools.memory_tool import MemoryStore
                     self._memory_store = MemoryStore(
@@ -1797,6 +1799,16 @@ class AIAgent:
         "If nothing is worth saving, just say 'Nothing to save.' and stop."
     )
 
+    
+    _PALACE_REVIEW_PROMPT = (
+        "Review the conversation above and archive a summary of the key events, "
+        "decisions, and outcomes to the MemPalace.\n\n"
+        "This ensures that future sessions can recall the context of this work "
+        "even after the current session history is lost.\n\n"
+        "Use the mempalace tool with action='record' and room='nexus' or 'workspace'. "
+        "Include a concise but informative summary of what was achieved."
+    )
+
     _COMBINED_REVIEW_PROMPT = (
         "Review the conversation above and consider two things:\n\n"
         "**Memory**: Has the user revealed things about themselves — their persona, "
@@ -1816,6 +1828,7 @@ class AIAgent:
         messages_snapshot: List[Dict],
         review_memory: bool = False,
         review_skills: bool = False,
+        review_palace: bool = False,
     ) -> None:
         """Spawn a background thread to review the conversation for memory/skill saves.
 
@@ -1826,13 +1839,20 @@ class AIAgent:
         """
         import threading
 
+
         # Pick the right prompt based on which triggers fired
-        if review_memory and review_skills:
+        prompts = []
-            prompt = self._COMBINED_REVIEW_PROMPT
+        if review_memory: prompts.append(self._MEMORY_REVIEW_PROMPT)
-        elif review_memory:
+        if review_skills: prompts.append(self._SKILL_REVIEW_PROMPT)
-            prompt = self._MEMORY_REVIEW_PROMPT
+        if review_palace: prompts.append(self._PALACE_REVIEW_PROMPT)
-        else:
+        
-            prompt = self._SKILL_REVIEW_PROMPT
+        if not prompts:
+            return
+            
+        prompt = "\n\n".join(prompts)
+        if len(prompts) > 1:
+            prompt = "Review the conversation above and perform the following tasks:\n\n" + prompt
+
 
         def _run_review():
             import contextlib, os as _os
@@ -6151,6 +6171,13 @@ class AIAgent:
             # Reset nudge counters
             if function_name == "memory":
                 self._turns_since_memory = 0
+
+        if (self._palace_nudge_interval > 0
+                and "mempalace" in self.valid_tool_names):
+            self._turns_since_palace += 1
+            if self._turns_since_palace >= self._palace_nudge_interval:
+                _should_review_palace = True
+                self._turns_since_palace = 0
             elif function_name == "skill_manage":
                 self._iters_since_skill = 0
 
@@ -7002,6 +7029,7 @@ class AIAgent:
         # Skill trigger is checked AFTER the agent loop completes, based on
         # how many tool iterations THIS turn used.
         _should_review_memory = False
+        _should_review_palace = False
         if (self._memory_nudge_interval > 0
                 and "memory" in self.valid_tool_names
                 and self._memory_store):
@@ -9155,12 +9183,13 @@ class AIAgent:
 
         # Background memory/skill review — runs AFTER the response is delivered
         # so it never competes with the user's task for model attention.
-        if final_response and not interrupted and (_should_review_memory or _should_review_skills):
+        if final_response and not interrupted and (_should_review_memory or _should_review_skills or _should_review_palace):
             try:
                 self._spawn_background_review(
                     messages_snapshot=list(messages),
                     review_memory=_should_review_memory,
                     review_skills=_should_review_skills,
+                    review_palace=_should_review_palace,
                 )
             except Exception:
                 pass  # Background review is best-effort

tools/mempalace_tool.py

@@ -0,0 +1,133 @@
+"""
+mempalace_tool.py — Interface for the MemPalace vector memory system.
+
+This tool allows the agent to search and record memories in a ChromaDB-backed
+vector database via the fleet_api.
+"""
+
+import json
+import os
+import requests
+import logging
+from typing import Dict, Any, List, Optional
+from tools.registry import registry
+
+logger = logging.getLogger(__name__)
+
+# Default to the fleet_api port defined in the-nexus
+MEMPALACE_API_URL = os.environ.get("MEMPALACE_API_URL", "http://127.0.0.1:7771")
+
+def mempalace_tool(
+    action: str,
+    query: str = "",
+    text: str = "",
+    room: Optional[str] = None,
+    wing: Optional[str] = None,
+    n_results: int = 10,
+    metadata: Optional[Dict[str, Any]] = None,
+) -> str:
+    """
+    Interface for the MemPalace vector memory system.
+    
+    Actions:
+        search: Search the palace for relevant memories.
+        record: Add a new memory to the palace.
+        wings: List available wizard wings.
+    """
+    try:
+        if action == "search":
+            if not query:
+                return json.dumps({"success": False, "error": "Query is required for search."})
+            
+            params = {"q": query, "n": n_results}
+            if room:
+                params["room"] = room
+            
+            response = requests.get(f"{MEMPALACE_API_URL}/search", params=params, timeout=10)
+            response.raise_for_status()
+            return json.dumps({"success": True, "data": response.json()})
+
+        elif action == "record":
+            if not text:
+                return json.dumps({"success": False, "error": "Text is required for record."})
+            
+            payload = {
+                "text": text,
+                "room": room or "general",
+                "wing": wing,
+                "metadata": metadata or {}
+            }
+            
+            response = requests.post(f"{MEMPALACE_API_URL}/record", json=payload, timeout=10)
+            response.raise_for_status()
+            return json.dumps({"success": True, "data": response.json()})
+
+        elif action == "wings":
+            response = requests.get(f"{MEMPALACE_API_URL}/wings", timeout=10)
+            response.raise_for_status()
+            return json.dumps({"success": True, "data": response.json()})
+
+        else:
+            return json.dumps({"success": False, "error": f"Unknown action: {action}"})
+
+    except requests.exceptions.ConnectionError:
+        return json.dumps({
+            "success": False, 
+            "error": f"Could not connect to MemPalace API at {MEMPALACE_API_URL}. Ensure fleet_api.py is running."
+        })
+    except Exception as e:
+        logger.error(f"MemPalace tool error: {e}")
+        return json.dumps({"success": False, "error": str(e)})
+
+MEMPALACE_SCHEMA = {
+    "name": "mempalace",
+    "description": (
+        "Search or record memories in the MemPalace vector database. "
+        "Use this for long-term, high-volume memory that exceeds the curated memory limits. "
+        "The palace is organized into 'rooms' (e.g., forge, hermes, nexus, issues, experiments)."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "action": {
+                "type": "string",
+                "enum": ["search", "record", "wings"],
+                "description": "The action to perform."
+            },
+            "query": {
+                "type": "string",
+                "description": "The search query (required for 'search')."
+            },
+            "text": {
+                "type": "string",
+                "description": "The memory text to record (required for 'record')."
+            },
+            "room": {
+                "type": "string",
+                "description": "Optional room filter or target room (e.g., forge, hermes, nexus, issues, experiments)."
+            },
+            "wing": {
+                "type": "string",
+                "description": "Optional wizard wing filter."
+            },
+            "n_results": {
+                "type": "integer",
+                "default": 10,
+                "description": "Maximum number of results to return."
+            },
+            "metadata": {
+                "type": "object",
+                "description": "Optional metadata for the memory (only for 'record')."
+            }
+        },
+        "required": ["action"]
+    }
+}
+
+registry.register(
+    name="mempalace",
+    toolset="mempalace",
+    schema=MEMPALACE_SCHEMA,
+    handler=lambda args, **kw: mempalace_tool(**args),
+    emoji="🏛️",
+)