diff --git a/src/brain/schema.py b/src/brain/schema.py
index 8175dff..d504ae9 100644
--- a/src/brain/schema.py
+++ b/src/brain/schema.py
@@ -7,9 +7,9 @@ SQL to initialize rqlite with memories and tasks tables.
 SCHEMA_VERSION = 1
 
 INIT_SQL = """
--- Enable SQLite extensions
-.load vector0
-.load vec0
+-- Note: sqlite-vec extensions must be loaded programmatically
+-- via conn.load_extension("vector0") / conn.load_extension("vec0")
+-- before executing this schema. Dot-commands are CLI-only.
 
 -- Memories table with vector search
 CREATE TABLE IF NOT EXISTS memories (
diff --git a/src/integrations/chat_bridge/vendors/discord.py b/src/integrations/chat_bridge/vendors/discord.py
index 54059b9..de155d7 100644
--- a/src/integrations/chat_bridge/vendors/discord.py
+++ b/src/integrations/chat_bridge/vendors/discord.py
@@ -32,6 +32,25 @@ logger = logging.getLogger(__name__)
 
 _STATE_FILE = Path(__file__).parent.parent.parent.parent / "discord_state.json"
 
+# Module-level agent singleton — reused across all Discord messages.
+# Mirrors the pattern from timmy.session._agent.
+_discord_agent = None
+
+
+def _get_discord_agent():
+    """Lazy-initialize the Discord agent singleton."""
+    global _discord_agent
+    if _discord_agent is None:
+        from timmy.agent import create_timmy
+
+        try:
+            _discord_agent = create_timmy()
+            logger.info("Discord: Timmy agent initialized (singleton)")
+        except Exception as exc:
+            logger.error("Discord: Failed to create Timmy agent: %s", exc)
+            raise
+    return _discord_agent
+
 
 class DiscordVendor(ChatPlatform):
     """Discord integration with native thread conversations.
@@ -330,17 +349,28 @@ class DiscordVendor(ChatPlatform):
         thread = await self._get_or_create_thread(message)
         target = thread or message.channel
 
-        # Run Timmy agent
-        try:
-            from timmy.agent import create_timmy
+        # Derive session_id for per-conversation history via Agno's SQLite
+        if thread:
+            session_id = f"discord_{thread.id}"
+        else:
+            session_id = f"discord_{message.channel.id}"
 
-            agent = create_timmy()
-            run = await asyncio.to_thread(agent.run, content, stream=False)
+        # Run Timmy agent (singleton, with session continuity)
+        try:
+            agent = _get_discord_agent()
+            run = await asyncio.to_thread(
+                agent.run, content, stream=False, session_id=session_id
+            )
             response = run.content if hasattr(run, "content") else str(run)
         except Exception as exc:
             logger.error("Timmy error in Discord handler: %s", exc)
             response = f"Timmy is offline: {exc}"
 
+        # Strip hallucinated tool-call JSON and chain-of-thought narration
+        from timmy.session import _clean_response
+
+        response = _clean_response(response)
+
         # Discord has a 2000 character limit
         for chunk in _chunk_message(response, 2000):
             await target.send(chunk)
diff --git a/src/timmy/agents/timmy.py b/src/timmy/agents/timmy.py
index 9ef2de2..2189a78 100644
--- a/src/timmy/agents/timmy.py
+++ b/src/timmy/agents/timmy.py
@@ -82,13 +82,10 @@ def build_timmy_context_sync() -> dict[str, Any]:
         logger.warning("Could not load agents for context: %s", exc)
         ctx["agents"] = []
     
-    # 3. Read hot memory
+    # 3. Read hot memory (via HotMemory to auto-create if missing)
     try:
-        memory_path = Path(settings.repo_root) / "MEMORY.md"
-        if memory_path.exists():
-            ctx["memory"] = memory_path.read_text()[:2000]  # First 2000 chars
-        else:
-            ctx["memory"] = "(MEMORY.md not found)"
+        from timmy.memory_system import memory_system
+        ctx["memory"] = memory_system.hot.read()[:2000]
     except Exception as exc:
         logger.warning("Could not load memory for context: %s", exc)
         ctx["memory"] = "(Memory unavailable)"
diff --git a/src/timmy/memory/vector_store.py b/src/timmy/memory/vector_store.py
index dbcadfb..43562ad 100644
--- a/src/timmy/memory/vector_store.py
+++ b/src/timmy/memory/vector_store.py
@@ -12,7 +12,7 @@ from datetime import datetime, timezone
 from pathlib import Path
 from typing import Optional
 
-DB_PATH = Path("data/swarm.db")
+DB_PATH = Path(__file__).parent.parent.parent.parent / "data" / "swarm.db"
 
 # Simple embedding function using sentence-transformers if available,
 # otherwise fall back to keyword-based "pseudo-embeddings"
diff --git a/src/timmy/memory_system.py b/src/timmy/memory_system.py
index 9cfe029..bbc7bff 100644
--- a/src/timmy/memory_system.py
+++ b/src/timmy/memory_system.py
@@ -431,8 +431,28 @@ class MemorySystem:
         return "\n".join(summary_parts) if summary_parts else ""
     
     def get_system_context(self) -> str:
-        """Get full context for system prompt injection."""
-        return self.start_session()
+        """Get full context for system prompt injection.
+
+        Unlike start_session(), this does NOT clear the handoff.
+        Safe to call multiple times without data loss.
+        """
+        context_parts = []
+
+        # 1. Hot memory
+        hot_content = self.hot.read()
+        context_parts.append("## Hot Memory\n" + hot_content)
+
+        # 2. Last session handoff (read-only, do NOT clear)
+        handoff_content = self.handoff.read_handoff()
+        if handoff_content:
+            context_parts.append("## Previous Session\n" + handoff_content)
+
+        # 3. User profile (key fields only)
+        profile = self._load_user_profile_summary()
+        if profile:
+            context_parts.append("## User Context\n" + profile)
+
+        return "\n\n---\n\n".join(context_parts)
 
 
 # Module-level singleton
diff --git a/src/timmy/tools_intro/__init__.py b/src/timmy/tools_intro/__init__.py
index 19e511d..07ec012 100644
--- a/src/timmy/tools_intro/__init__.py
+++ b/src/timmy/tools_intro/__init__.py
@@ -164,10 +164,10 @@ def get_memory_status() -> dict[str, Any]:
         if sem_db.exists():
             conn = sqlite3.connect(str(sem_db))
             row = conn.execute(
-                "SELECT COUNT(*) FROM sqlite_master WHERE type='table' AND name='vectors'"
+                "SELECT COUNT(*) FROM sqlite_master WHERE type='table' AND name='chunks'"
             ).fetchone()
             if row and row[0]:
-                count = conn.execute("SELECT COUNT(*) FROM vectors").fetchone()
+                count = conn.execute("SELECT COUNT(*) FROM chunks").fetchone()
                 tier3_info["available"] = True
                 tier3_info["vector_count"] = count[0] if count else 0
             conn.close()
diff --git a/tests/timmy/test_agents_timmy.py b/tests/timmy/test_agents_timmy.py
index 5f63877..ffae023 100644
--- a/tests/timmy/test_agents_timmy.py
+++ b/tests/timmy/test_agents_timmy.py
@@ -63,8 +63,17 @@ class TestBuildContext:
         memory_file.write_text("# Important memories\nRemember this.")
         mock_settings.repo_root = str(tmp_path)
 
-        ctx = build_timmy_context_sync()
-        assert "Important memories" in ctx["memory"]
+        # Patch HotMemory path so it reads from tmp_path
+        from timmy.memory_system import memory_system
+        original_path = memory_system.hot.path
+        memory_system.hot.path = memory_file
+        memory_system.hot._content = None  # Clear cache
+        try:
+            ctx = build_timmy_context_sync()
+            assert "Important memories" in ctx["memory"]
+        finally:
+            memory_system.hot.path = original_path
+            memory_system.hot._content = None
 
 
 class TestFormatPrompt: