feat: cross-session agent memory via MemPalace (#1124)

Integrates MemPalace for persistent agent memory across sessions.
Agents recall context at session start, store important decisions,
and write diary entries at session end.

## What's added

agent/memory.py — AgentMemory class:
  - recall_context(): Load L0/L1 context (diaries, facts, relevant memories)
  - remember(): Store decisions and facts by room
  - write_diary(): Auto-generate session summary from transcript
  - start_session/end_session(): Session lifecycle management
  - Graceful degradation when MemPalace unavailable

agent/memory_hooks.py — Drop-in session lifecycle hooks:
  - on_session_start(): Load context, return prompt block
  - on_user_turn/on_agent_turn/on_tool_call(): Record transcript
  - on_important_decision(): Store key decisions for long-term memory
  - on_session_end(): Write diary, clean up

bin/memory_mine.py — Mine session transcripts into MemPalace:
  - Parse JSONL session files
  - Generate compact summaries
  - Batch mining with --days filter
  - Dry run mode

tests/test_agent_memory.py — 31 tests covering:
  - SessionTranscript (create, turns, truncation, summary)
  - MemoryContext (empty, loaded, prompt formatting)
  - AgentMemory (create, factory, graceful degradation, lifecycle)
  - MemoryHooks (full lifecycle, before/after session guards)
  - Session mining (parse, summarize, find files, dry run)
  - Full lifecycle integration test

## Usage

.gitea/workflows/deploy.yml

@@ -27,10 +27,8 @@ jobs:
           username: ${{ secrets.DEPLOY_USER }}
           key: ${{ secrets.DEPLOY_SSH_KEY }}
           script: |
-            cd ~/the-nexus || git clone https://forge.alexanderwhitestone.com/Timmy_Foundation/the-nexus.git ~/the-nexus
+            cd ~/the-nexus || git clone http://143.198.27.163:3000/Timmy_Foundation/the-nexus.git ~/the-nexus
             cd ~/the-nexus
             git fetch origin main
             git reset --hard origin/main
-            docker compose build nexus-main
-            docker compose up -d --force-recreate nexus-main
-            echo "Nexus deployed — HTTP :8080, WS :8765"
+            ./deploy.sh main

Dockerfile

@@ -11,16 +11,11 @@ COPY nexus/ nexus/
 COPY server.py ./
 
 # Frontend assets referenced by index.html
-COPY index.html help.html style.css app.js boot.js bootstrap.mjs gofai_worker.js mempalace.js service-worker.js manifest.json ./
+COPY index.html help.html style.css app.js service-worker.js manifest.json ./
 
 # Config/data
 COPY portals.json vision.json robots.txt ./
 
-# Icons
-COPY icons/ icons/
-
-# Expose HTTP (static) and WebSocket
-EXPOSE 8080
 EXPOSE 8765
 
 CMD ["python3", "server.py"]

agent/__init__.py

@@ -0,0 +1,21 @@
+"""
+agent — Cross-session agent memory and lifecycle hooks.
+
+Provides persistent memory for agents via MemPalace integration.
+Agents recall context at session start and write diary entries at session end.
+
+Modules:
+    memory.py       — AgentMemory class (recall, remember, diary)
+    memory_hooks.py — Session lifecycle hooks (drop-in integration)
+"""
+
+from agent.memory import AgentMemory, MemoryContext, SessionTranscript, create_agent_memory
+from agent.memory_hooks import MemoryHooks
+
+__all__ = [
+    "AgentMemory",
+    "MemoryContext",
+    "MemoryHooks",
+    "SessionTranscript",
+    "create_agent_memory",
+]

agent/memory.py

@@ -0,0 +1,396 @@
+"""
+agent.memory — Cross-session agent memory via MemPalace.
+
+Gives agents persistent memory across sessions. On wake-up, agents
+recall relevant context from past sessions. On session end, they
+write a diary entry summarizing what happened.
+
+Architecture:
+    Session Start  → memory.recall_context() → inject L0/L1 into prompt
+    During Session → memory.remember()        → store important facts
+    Session End    → memory.write_diary()     → summarize session
+
+All operations degrade gracefully — if MemPalace is unavailable,
+the agent continues without memory and logs a warning.
+
+Usage:
+    from agent.memory import AgentMemory
+
+    mem = AgentMemory(agent_name="bezalel", wing="wing_bezalel")
+
+    # Session start — load context
+    context = mem.recall_context("What was I working on last time?")
+
+    # During session — store important decisions
+    mem.remember("Switched CI runner from GitHub Actions to self-hosted", room="forge")
+
+    # Session end — write diary
+    mem.write_diary("Fixed PR #1386, reconciled fleet registry locations")
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+import time
+from dataclasses import dataclass, field
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Optional
+
+logger = logging.getLogger("agent.memory")
+
+
+@dataclass
+class MemoryContext:
+    """Context loaded at session start from MemPalace."""
+    relevant_memories: list[dict] = field(default_factory=list)
+    recent_diaries: list[dict] = field(default_factory=list)
+    facts: list[dict] = field(default_factory=list)
+    loaded: bool = False
+    error: Optional[str] = None
+
+    def to_prompt_block(self) -> str:
+        """Format context as a text block to inject into the agent prompt."""
+        if not self.loaded:
+            return ""
+
+        parts = []
+
+        if self.recent_diaries:
+            parts.append("=== Recent Session Summaries ===")
+            for d in self.recent_diaries[:3]:
+                ts = d.get("timestamp", "")
+                text = d.get("text", "")
+                parts.append(f"[{ts}] {text[:500]}")
+
+        if self.facts:
+            parts.append("\n=== Known Facts ===")
+            for f in self.facts[:10]:
+                text = f.get("text", "")
+                parts.append(f"- {text[:200]}")
+
+        if self.relevant_memories:
+            parts.append("\n=== Relevant Past Memories ===")
+            for m in self.relevant_memories[:5]:
+                text = m.get("text", "")
+                score = m.get("score", 0)
+                parts.append(f"[{score:.2f}] {text[:300]}")
+
+        if not parts:
+            return ""
+
+        return "\n".join(parts)
+
+
+@dataclass
+class SessionTranscript:
+    """A running log of the current session for diary writing."""
+    agent_name: str
+    wing: str
+    started_at: str = field(
+        default_factory=lambda: datetime.now(timezone.utc).isoformat()
+    )
+    entries: list[dict] = field(default_factory=list)
+
+    def add_user_turn(self, text: str):
+        self.entries.append({
+            "role": "user",
+            "text": text[:2000],
+            "ts": time.time(),
+        })
+
+    def add_agent_turn(self, text: str):
+        self.entries.append({
+            "role": "agent",
+            "text": text[:2000],
+            "ts": time.time(),
+        })
+
+    def add_tool_call(self, tool: str, args: str, result_summary: str):
+        self.entries.append({
+            "role": "tool",
+            "tool": tool,
+            "args": args[:500],
+            "result": result_summary[:500],
+            "ts": time.time(),
+        })
+
+    def summary(self) -> str:
+        """Generate a compact transcript summary."""
+        if not self.entries:
+            return "Empty session."
+
+        turns = []
+        for e in self.entries[-20:]:  # last 20 entries
+            role = e["role"]
+            if role == "user":
+                turns.append(f"USER: {e['text'][:200]}")
+            elif role == "agent":
+                turns.append(f"AGENT: {e['text'][:200]}")
+            elif role == "tool":
+                turns.append(f"TOOL({e.get('tool','')}): {e.get('result','')[:150]}")
+
+        return "\n".join(turns)
+
+
+class AgentMemory:
+    """
+    Cross-session memory for an agent.
+
+    Wraps MemPalace with agent-specific conventions:
+    - Each agent has a wing (e.g., "wing_bezalel")
+    - Session summaries go in the "hermes" room
+    - Important decisions go in room-specific closets
+    - Facts go in the "nexus" room
+    """
+
+    def __init__(
+        self,
+        agent_name: str,
+        wing: Optional[str] = None,
+        palace_path: Optional[Path] = None,
+    ):
+        self.agent_name = agent_name
+        self.wing = wing or f"wing_{agent_name}"
+        self.palace_path = palace_path
+        self._transcript: Optional[SessionTranscript] = None
+        self._available: Optional[bool] = None
+
+    def _check_available(self) -> bool:
+        """Check if MemPalace is accessible."""
+        if self._available is not None:
+            return self._available
+
+        try:
+            from nexus.mempalace.searcher import search_memories, add_memory, _get_client
+            from nexus.mempalace.config import MEMPALACE_PATH
+
+            path = self.palace_path or MEMPALACE_PATH
+            _get_client(path)
+            self._available = True
+            logger.info(f"MemPalace available at {path}")
+        except Exception as e:
+            self._available = False
+            logger.warning(f"MemPalace unavailable: {e}")
+
+        return self._available
+
+    def recall_context(
+        self,
+        query: Optional[str] = None,
+        n_results: int = 5,
+    ) -> MemoryContext:
+        """
+        Load relevant context from past sessions.
+
+        Called at session start to inject L0/L1 memory into the prompt.
+
+        Args:
+            query: What to search for. If None, loads recent diary entries.
+            n_results: Max memories to recall.
+        """
+        ctx = MemoryContext()
+
+        if not self._check_available():
+            ctx.error = "MemPalace unavailable"
+            return ctx
+
+        try:
+            from nexus.mempalace.searcher import search_memories
+
+            # Load recent diary entries (session summaries)
+            ctx.recent_diaries = [
+                {"text": r.text, "score": r.score, "timestamp": r.metadata.get("timestamp", "")}
+                for r in search_memories(
+                    "session summary",
+                    palace_path=self.palace_path,
+                    wing=self.wing,
+                    room="hermes",
+                    n_results=3,
+                )
+            ]
+
+            # Load known facts
+            ctx.facts = [
+                {"text": r.text, "score": r.score}
+                for r in search_memories(
+                    "important facts decisions",
+                    palace_path=self.palace_path,
+                    wing=self.wing,
+                    room="nexus",
+                    n_results=5,
+                )
+            ]
+
+            # Search for relevant memories if query provided
+            if query:
+                ctx.relevant_memories = [
+                    {"text": r.text, "score": r.score, "room": r.room}
+                    for r in search_memories(
+                        query,
+                        palace_path=self.palace_path,
+                        wing=self.wing,
+                        n_results=n_results,
+                    )
+                ]
+
+            ctx.loaded = True
+
+        except Exception as e:
+            ctx.error = str(e)
+            logger.warning(f"Failed to recall context: {e}")
+
+        return ctx
+
+    def remember(
+        self,
+        text: str,
+        room: str = "nexus",
+        source_file: str = "",
+        metadata: Optional[dict] = None,
+    ) -> Optional[str]:
+        """
+        Store a memory.
+
+        Args:
+            text: The memory content.
+            room: Target room (forge, hermes, nexus, issues, experiments).
+            source_file: Optional source attribution.
+            metadata: Extra metadata.
+
+        Returns:
+            Document ID if stored, None if MemPalace unavailable.
+        """
+        if not self._check_available():
+            logger.warning("Cannot store memory — MemPalace unavailable")
+            return None
+
+        try:
+            from nexus.mempalace.searcher import add_memory
+
+            doc_id = add_memory(
+                text=text,
+                room=room,
+                wing=self.wing,
+                palace_path=self.palace_path,
+                source_file=source_file,
+                extra_metadata=metadata or {},
+            )
+            logger.debug(f"Stored memory in {room}: {text[:80]}...")
+            return doc_id
+
+        except Exception as e:
+            logger.warning(f"Failed to store memory: {e}")
+            return None
+
+    def write_diary(
+        self,
+        summary: Optional[str] = None,
+    ) -> Optional[str]:
+        """
+        Write a session diary entry to MemPalace.
+
+        Called at session end. If summary is None, auto-generates one
+        from the session transcript.
+
+        Args:
+            summary: Override summary text. If None, generates from transcript.
+
+        Returns:
+            Document ID if stored, None if unavailable.
+        """
+        if summary is None and self._transcript:
+            summary = self._transcript.summary()
+
+        if not summary:
+            return None
+
+        timestamp = datetime.now(timezone.utc).isoformat()
+        diary_text = f"[{timestamp}] Session by {self.agent_name}:\n{summary}"
+
+        return self.remember(
+            diary_text,
+            room="hermes",
+            metadata={
+                "type": "session_diary",
+                "agent": self.agent_name,
+                "timestamp": timestamp,
+                "entry_count": len(self._transcript.entries) if self._transcript else 0,
+            },
+        )
+
+    def start_session(self) -> SessionTranscript:
+        """
+        Begin a new session transcript.
+
+        Returns the transcript object for recording turns.
+        """
+        self._transcript = SessionTranscript(
+            agent_name=self.agent_name,
+            wing=self.wing,
+        )
+        logger.info(f"Session started for {self.agent_name}")
+        return self._transcript
+
+    def end_session(self, diary_summary: Optional[str] = None) -> Optional[str]:
+        """
+        End the current session, write diary, return diary doc ID.
+        """
+        doc_id = self.write_diary(diary_summary)
+        self._transcript = None
+        logger.info(f"Session ended for {self.agent_name}")
+        return doc_id
+
+    def search(
+        self,
+        query: str,
+        room: Optional[str] = None,
+        n_results: int = 5,
+    ) -> list[dict]:
+        """
+        Search memories. Useful during a session for recall.
+
+        Returns list of {text, room, wing, score} dicts.
+        """
+        if not self._check_available():
+            return []
+
+        try:
+            from nexus.mempalace.searcher import search_memories
+
+            results = search_memories(
+                query,
+                palace_path=self.palace_path,
+                wing=self.wing,
+                room=room,
+                n_results=n_results,
+            )
+            return [
+                {"text": r.text, "room": r.room, "wing": r.wing, "score": r.score}
+                for r in results
+            ]
+
+        except Exception as e:
+            logger.warning(f"Search failed: {e}")
+            return []
+
+
+# --- Fleet-wide memory helpers ---
+
+def create_agent_memory(
+    agent_name: str,
+    palace_path: Optional[Path] = None,
+) -> AgentMemory:
+    """
+    Factory for creating AgentMemory with standard config.
+
+    Reads wing from MEMPALACE_WING env or defaults to wing_{agent_name}.
+    """
+    wing = os.environ.get("MEMPALACE_WING", f"wing_{agent_name}")
+    return AgentMemory(
+        agent_name=agent_name,
+        wing=wing,
+        palace_path=palace_path,
+    )

agent/memory_hooks.py

@@ -0,0 +1,183 @@
+"""
+agent.memory_hooks — Session lifecycle hooks for agent memory.
+
+Integrates AgentMemory into the agent session lifecycle:
+- on_session_start: Load context, inject into prompt
+- on_user_turn: Record user input
+- on_agent_turn: Record agent output
+- on_tool_call: Record tool usage
+- on_session_end: Write diary, clean up
+
+These hooks are designed to be called from the Hermes harness or
+any agent framework. They're fire-and-forget — failures are logged
+but never crash the session.
+
+Usage:
+    from agent.memory_hooks import MemoryHooks
+
+    hooks = MemoryHooks(agent_name="bezalel")
+    hooks.on_session_start()  # loads context
+
+    # In your agent loop:
+    hooks.on_user_turn("Check CI pipeline health")
+    hooks.on_agent_turn("Running CI check...")
+    hooks.on_tool_call("shell", "pytest tests/", "12 passed")
+
+    # End of session:
+    hooks.on_session_end()  # writes diary
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Optional
+
+from agent.memory import AgentMemory, MemoryContext, create_agent_memory
+
+logger = logging.getLogger("agent.memory_hooks")
+
+
+class MemoryHooks:
+    """
+    Drop-in session lifecycle hooks for agent memory.
+
+    Wraps AgentMemory with error boundaries — every hook catches
+    exceptions and logs warnings so memory failures never crash
+    the agent session.
+    """
+
+    def __init__(
+        self,
+        agent_name: str,
+        palace_path=None,
+        auto_diary: bool = True,
+    ):
+        self.agent_name = agent_name
+        self.auto_diary = auto_diary
+        self._memory: Optional[AgentMemory] = None
+        self._context: Optional[MemoryContext] = None
+        self._active = False
+
+    @property
+    def memory(self) -> AgentMemory:
+        if self._memory is None:
+            self._memory = create_agent_memory(
+                self.agent_name,
+                palace_path=getattr(self, '_palace_path', None),
+            )
+        return self._memory
+
+    def on_session_start(self, query: Optional[str] = None) -> str:
+        """
+        Called at session start. Loads context from MemPalace.
+
+        Returns a prompt block to inject into the agent's context, or
+        empty string if memory is unavailable.
+
+        Args:
+            query: Optional recall query (e.g., "What was I working on?")
+        """
+        try:
+            self.memory.start_session()
+            self._active = True
+
+            self._context = self.memory.recall_context(query=query)
+            block = self._context.to_prompt_block()
+
+            if block:
+                logger.info(
+                    f"Loaded {len(self._context.recent_diaries)} diaries, "
+                    f"{len(self._context.facts)} facts, "
+                    f"{len(self._context.relevant_memories)} relevant memories "
+                    f"for {self.agent_name}"
+                )
+            else:
+                logger.info(f"No prior memory for {self.agent_name}")
+
+            return block
+
+        except Exception as e:
+            logger.warning(f"Session start memory hook failed: {e}")
+            return ""
+
+    def on_user_turn(self, text: str):
+        """Record a user message."""
+        if not self._active:
+            return
+        try:
+            if self.memory._transcript:
+                self.memory._transcript.add_user_turn(text)
+        except Exception as e:
+            logger.debug(f"Failed to record user turn: {e}")
+
+    def on_agent_turn(self, text: str):
+        """Record an agent response."""
+        if not self._active:
+            return
+        try:
+            if self.memory._transcript:
+                self.memory._transcript.add_agent_turn(text)
+        except Exception as e:
+            logger.debug(f"Failed to record agent turn: {e}")
+
+    def on_tool_call(self, tool: str, args: str, result_summary: str):
+        """Record a tool invocation."""
+        if not self._active:
+            return
+        try:
+            if self.memory._transcript:
+                self.memory._transcript.add_tool_call(tool, args, result_summary)
+        except Exception as e:
+            logger.debug(f"Failed to record tool call: {e}")
+
+    def on_important_decision(self, text: str, room: str = "nexus"):
+        """
+        Record an important decision or fact for long-term memory.
+
+        Use this when the agent makes a significant decision that
+        should persist beyond the current session.
+        """
+        try:
+            self.memory.remember(text, room=room, metadata={"type": "decision"})
+            logger.info(f"Remembered decision: {text[:80]}...")
+        except Exception as e:
+            logger.warning(f"Failed to remember decision: {e}")
+
+    def on_session_end(self, summary: Optional[str] = None) -> Optional[str]:
+        """
+        Called at session end. Writes diary entry.
+
+        Args:
+            summary: Override diary text. If None, auto-generates.
+
+        Returns:
+            Diary document ID, or None.
+        """
+        if not self._active:
+            return None
+
+        try:
+            doc_id = self.memory.end_session(diary_summary=summary)
+            self._active = False
+            self._context = None
+            return doc_id
+        except Exception as e:
+            logger.warning(f"Session end memory hook failed: {e}")
+            self._active = False
+            return None
+
+    def search(self, query: str, room: Optional[str] = None) -> list[dict]:
+        """
+        Search memories during a session.
+
+        Returns list of {text, room, wing, score}.
+        """
+        try:
+            return self.memory.search(query, room=room)
+        except Exception as e:
+            logger.warning(f"Memory search failed: {e}")
+            return []
+
+    @property
+    def is_active(self) -> bool:
+        return self._active

app.js

@@ -2188,11 +2188,7 @@ function connectHermes() {
   }
 
   const protocol = window.location.protocol === 'https:' ? 'wss:' : 'ws:';
-  // WS gateway runs on HTTP port - 115 (8080->8765, 8081->8766)
-  const wsHost = window.location.hostname;
-  const httpPort = parseInt(window.location.port) || 8080;
-  const wsPort = httpPort === 8081 ? 8766 : 8765;
-  const wsUrl = `${protocol}//${wsHost}:${wsPort}/api/world/ws`;
+  const wsUrl = `${protocol}//${window.location.host}/api/world/ws`;
   
   console.log(`Connecting to Hermes at ${wsUrl}...`);
   hermesWs = new WebSocket(wsUrl);

bin/memory_mine.py

@@ -0,0 +1,258 @@
+#!/usr/bin/env python3
+"""
+memory_mine.py — Mine session transcripts into MemPalace.
+
+Reads Hermes session logs (JSONL format) and stores summaries
+in the palace. Supports batch mining, single-file processing,
+and live directory watching.
+
+Usage:
+    # Mine a single session file
+    python3 bin/memory_mine.py ~/.hermes/sessions/2026-04-13.jsonl
+
+    # Mine all sessions from last 7 days
+    python3 bin/memory_mine.py --days 7
+
+    # Mine a specific wing's sessions
+    python3 bin/memory_mine.py --wing wing_bezalel --days 14
+
+    # Dry run — show what would be mined
+    python3 bin/memory_mine.py --dry-run --days 7
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import logging
+import os
+import sys
+import time
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+from typing import Optional
+
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s [%(levelname)s] %(message)s",
+    datefmt="%Y-%m-%d %H:%M:%S",
+)
+logger = logging.getLogger("memory-mine")
+
+REPO_ROOT = Path(__file__).resolve().parent.parent
+if str(REPO_ROOT) not in sys.path:
+    sys.path.insert(0, str(REPO_ROOT))
+
+
+def parse_session_file(path: Path) -> list[dict]:
+    """
+    Parse a JSONL session file into turns.
+
+    Each line is expected to be a JSON object with:
+    - role: "user" | "assistant" | "system" | "tool"
+    - content: text
+    - timestamp: ISO string (optional)
+    """
+    turns = []
+    with open(path) as f:
+        for i, line in enumerate(f):
+            line = line.strip()
+            if not line:
+                continue
+            try:
+                turn = json.loads(line)
+                turns.append(turn)
+            except json.JSONDecodeError:
+                logger.debug(f"Skipping malformed line {i+1} in {path}")
+    return turns
+
+
+def summarize_session(turns: list[dict], agent_name: str = "unknown") -> str:
+    """
+    Generate a compact summary of a session's turns.
+
+    Keeps user messages and key agent responses, strips noise.
+    """
+    if not turns:
+        return "Empty session."
+
+    user_msgs = []
+    agent_msgs = []
+    tool_calls = []
+
+    for turn in turns:
+        role = turn.get("role", "")
+        content = str(turn.get("content", ""))[:300]
+
+        if role == "user":
+            user_msgs.append(content)
+        elif role == "assistant":
+            agent_msgs.append(content)
+        elif role == "tool":
+            tool_name = turn.get("name", turn.get("tool", "unknown"))
+            tool_calls.append(f"{tool_name}: {content[:150]}")
+
+    parts = [f"Session by {agent_name}:"]
+
+    if user_msgs:
+        parts.append(f"\nUser asked ({len(user_msgs)} messages):")
+        for msg in user_msgs[:5]:
+            parts.append(f"  - {msg[:200]}")
+        if len(user_msgs) > 5:
+            parts.append(f"  ... and {len(user_msgs) - 5} more")
+
+    if agent_msgs:
+        parts.append(f"\nAgent responded ({len(agent_msgs)} messages):")
+        for msg in agent_msgs[:3]:
+            parts.append(f"  - {msg[:200]}")
+
+    if tool_calls:
+        parts.append(f"\nTools used ({len(tool_calls)} calls):")
+        for tc in tool_calls[:5]:
+            parts.append(f"  - {tc}")
+
+    return "\n".join(parts)
+
+
+def mine_session(
+    path: Path,
+    wing: str,
+    palace_path: Optional[Path] = None,
+    dry_run: bool = False,
+) -> Optional[str]:
+    """
+    Mine a single session file into MemPalace.
+
+    Returns the document ID if stored, None on failure or dry run.
+    """
+    try:
+        from agent.memory import AgentMemory
+    except ImportError:
+        logger.error("Cannot import agent.memory — is the repo in PYTHONPATH?")
+        return None
+
+    turns = parse_session_file(path)
+    if not turns:
+        logger.debug(f"Empty session file: {path}")
+        return None
+
+    agent_name = wing.replace("wing_", "")
+    summary = summarize_session(turns, agent_name)
+
+    if dry_run:
+        print(f"\n--- {path.name} ---")
+        print(summary[:500])
+        print(f"({len(turns)} turns)")
+        return None
+
+    mem = AgentMemory(agent_name=agent_name, wing=wing, palace_path=palace_path)
+    doc_id = mem.remember(
+        summary,
+        room="hermes",
+        source_file=str(path),
+        metadata={
+            "type": "mined_session",
+            "source": str(path),
+            "turn_count": len(turns),
+            "agent": agent_name,
+            "timestamp": datetime.now(timezone.utc).isoformat(),
+        },
+    )
+
+    if doc_id:
+        logger.info(f"Mined {path.name} → {doc_id} ({len(turns)} turns)")
+    else:
+        logger.warning(f"Failed to mine {path.name}")
+
+    return doc_id
+
+
+def find_session_files(
+    sessions_dir: Path,
+    days: int = 7,
+    pattern: str = "*.jsonl",
+) -> list[Path]:
+    """
+    Find session files from the last N days.
+    """
+    cutoff = datetime.now() - timedelta(days=days)
+    files = []
+
+    if not sessions_dir.exists():
+        logger.warning(f"Sessions directory not found: {sessions_dir}")
+        return files
+
+    for path in sorted(sessions_dir.glob(pattern)):
+        # Use file modification time as proxy for session date
+        mtime = datetime.fromtimestamp(path.stat().st_mtime)
+        if mtime >= cutoff:
+            files.append(path)
+
+    return files
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(
+        description="Mine session transcripts into MemPalace"
+    )
+    parser.add_argument(
+        "files", nargs="*", help="Session files to mine (JSONL format)"
+    )
+    parser.add_argument(
+        "--days", type=int, default=7,
+        help="Mine sessions from last N days (default: 7)"
+    )
+    parser.add_argument(
+        "--sessions-dir",
+        default=str(Path.home() / ".hermes" / "sessions"),
+        help="Directory containing session JSONL files"
+    )
+    parser.add_argument(
+        "--wing", default=None,
+        help="Wing name (default: auto-detect from MEMPALACE_WING env or 'wing_timmy')"
+    )
+    parser.add_argument(
+        "--palace-path", default=None,
+        help="Override palace path"
+    )
+    parser.add_argument(
+        "--dry-run", action="store_true",
+        help="Show what would be mined without storing"
+    )
+
+    args = parser.parse_args(argv)
+
+    wing = args.wing or os.environ.get("MEMPALACE_WING", "wing_timmy")
+    palace_path = Path(args.palace_path) if args.palace_path else None
+
+    if args.files:
+        files = [Path(f) for f in args.files]
+    else:
+        sessions_dir = Path(args.sessions_dir)
+        files = find_session_files(sessions_dir, days=args.days)
+
+    if not files:
+        logger.info("No session files found to mine.")
+        return 0
+
+    logger.info(f"Mining {len(files)} session files (wing={wing})")
+
+    mined = 0
+    failed = 0
+    for path in files:
+        result = mine_session(path, wing=wing, palace_path=palace_path, dry_run=args.dry_run)
+        if result:
+            mined += 1
+        elif result is None and not args.dry_run:
+            failed += 1
+
+    if args.dry_run:
+        logger.info(f"Dry run complete — {len(files)} files would be mined")
+    else:
+        logger.info(f"Mining complete — {mined} mined, {failed} failed")
+
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

deploy.sh

@@ -1,7 +1,7 @@
 #!/usr/bin/env bash
 # deploy.sh — spin up (or update) the Nexus staging environment
-# Usage: ./deploy.sh          — rebuild and restart nexus-main (HTTP :8080, WS :8765)
-#        ./deploy.sh staging  — rebuild and restart nexus-staging (HTTP :8081, WS :8766)
+# Usage: ./deploy.sh          — rebuild and restart nexus-main (port 4200)
+#        ./deploy.sh staging  — rebuild and restart nexus-staging (port 4201)
 set -euo pipefail
 
 SERVICE="${1:-nexus-main}"
@@ -14,12 +14,4 @@ esac
 echo "==> Deploying $SERVICE …"
 docker compose build "$SERVICE"
 docker compose up -d --force-recreate "$SERVICE"
-
-if [ "$SERVICE" = "nexus-main" ]; then
-  echo "==> HTTP: http://localhost:8080"
-  echo "==> WS:   ws://localhost:8765"
-else
-  echo "==> HTTP: http://localhost:8081"
-  echo "==> WS:   ws://localhost:8766"
-fi
 echo "==> Done. Container: $SERVICE"

docker-compose.yml

@@ -6,12 +6,10 @@ services:
     container_name: nexus-main
     restart: unless-stopped
     ports:
-      - "8080:8080"
       - "8765:8765"
   nexus-staging:
     build: .
     container_name: nexus-staging
     restart: unless-stopped
     ports:
-      - "8081:8080"
-      - "8766:8765"
+      - "8766:8765"

run.sh

@@ -1,16 +0,0 @@
-#!/usr/bin/env bash
-# run.sh — run Nexus locally without Docker
-# Usage: ./run.sh              — HTTP :8080, WS :8765
-#        NEXUS_HTTP_PORT=9090 ./run.sh  — custom HTTP port
-set -euo pipefail
-
-cd "$(dirname "$0")"
-
-# Install deps if missing
-if ! python3 -c "import websockets" 2>/dev/null; then
-  echo "==> Installing dependencies..."
-  pip3 install -r requirements.txt
-fi
-
-echo "==> Starting Nexus server..."
-exec python3 server.py

server.py

@@ -1,190 +1,92 @@
 #!/usr/bin/env python3
 """
-The Nexus — Unified HTTP + WebSocket server.
-
-Serves static frontend files (Three.js app) over HTTP on port 8080
-and runs the WebSocket gateway on port 8765.
-Single-process, single-command deployment — no nginx required.
+The Nexus WebSocket Gateway — Robust broadcast bridge for Timmy's consciousness.
+This server acts as the central hub for the-nexus, connecting the mind (nexus_think.py),
+the body (Evennia/Morrowind), and the visualization surface.
 """
 import asyncio
 import json
 import logging
-import mimetypes
-import os
 import signal
 import sys
-from http import HTTPStatus
-from pathlib import Path
 from typing import Set
 
+# Branch protected file - see POLICY.md
 import websockets
-import websockets.asyncio.server
 
-# ---------------------------------------------------------------------------
 # Configuration
-# ---------------------------------------------------------------------------
-HTTP_PORT = int(os.environ.get("NEXUS_HTTP_PORT", "8080"))
-WS_PORT = int(os.environ.get("NEXUS_WS_PORT", "8765"))
-HOST = os.environ.get("NEXUS_HOST", "0.0.0.0")
-ROOT = Path(__file__).resolve().parent
+PORT = 8765
+HOST = "0.0.0.0"  # Allow external connections if needed
 
-# Static file extensions we're willing to serve
-SAFE_SUFFIXES = {
-    ".html", ".htm", ".css", ".js", ".mjs", ".json",
-    ".png", ".jpg", ".jpeg", ".gif", ".svg", ".ico",
-    ".woff", ".woff2", ".ttf", ".eot",
-    ".txt", ".xml", ".webmanifest",
-}
-
-# ---------------------------------------------------------------------------
-# Logging
-# ---------------------------------------------------------------------------
+# Logging setup
 logging.basicConfig(
     level=logging.INFO,
-    format="%(asctime)s [%(levelname)s] %(message)s",
-    datefmt="%Y-%m-%d %H:%M:%S",
+    format='%(asctime)s [%(levelname)s] %(message)s',
+    datefmt='%Y-%m-%d %H:%M:%S'
 )
-log = logging.getLogger("nexus")
+logger = logging.getLogger("nexus-gateway")
 
-# ---------------------------------------------------------------------------
-# HTTP — static file server
-# ---------------------------------------------------------------------------
-async def http_handler(reader: asyncio.StreamReader, writer: asyncio.StreamWriter):
-    """Minimal async HTTP/1.1 server for static files."""
-    try:
-        # Read request line + headers (max 8 KB)
-        data = b""
-        while b"\r\n\r\n" not in data and len(data) < 8192:
-            chunk = await asyncio.wait_for(reader.read(4096), timeout=10)
-            if not chunk:
-                return
-            data += chunk
-        header_text = data.split(b"\r\n\r\n", 1)[0].decode("utf-8", errors="replace")
-        lines = header_text.split("\r\n")
-        request_line = lines[0]
-        parts = request_line.split(" ", 2)
-        if len(parts) < 2:
-            writer.close()
-            return
-        method, raw_path = parts[0], parts[1]
+# State
+clients: Set[websockets.WebSocketServerProtocol] = set()
 
-        if method not in ("GET", "HEAD"):
-            _write_response(writer, HTTPStatus.METHOD_NOT_ALLOWED, b"Method Not Allowed")
-            return
-
-        # Normalise path — prevent directory traversal
-        safe_path = raw_path.split("?", 1)[0].split("#", 1)[0]
-        safe_path = os.path.normpath(safe_path).lstrip("/")
-        if not safe_path:
-            safe_path = "index.html"
-
-        file_path = ROOT / safe_path
-        # Reject traversal
-        if not str(file_path.resolve()).startswith(str(ROOT)):
-            _write_response(writer, HTTPStatus.FORBIDDEN, b"Forbidden")
-            return
-
-        if not file_path.exists() or file_path.is_dir():
-            # Try index.html for directories
-            if file_path.is_dir() and (file_path / "index.html").exists():
-                file_path = file_path / "index.html"
-            else:
-                _write_response(writer, HTTPStatus.NOT_FOUND, b"Not Found")
-                return
-
-        suffix = file_path.suffix.lower()
-        if suffix not in SAFE_SUFFIXES:
-            _write_response(writer, HTTPStatus.FORBIDDEN, b"Forbidden")
-            return
-
-        content_type = mimetypes.guess_type(str(file_path))[0] or "application/octet-stream"
-        body = file_path.read_bytes()
-        headers = {
-            "Content-Type": content_type,
-            "Content-Length": str(len(body)),
-            "Cache-Control": "no-cache",
-            "Access-Control-Allow-Origin": "*",
-        }
-        _write_response(writer, HTTPStatus.OK, body, headers, method == "HEAD")
-    except (asyncio.TimeoutError, ConnectionError):
-        pass
-    except Exception as exc:
-        log.warning("HTTP handler error: %s", exc)
-    finally:
-        try:
-            writer.close()
-            await writer.wait_closed()
-        except Exception:
-            pass
-
-
-def _write_response(writer, status, body, headers=None, head_only=False):
-    line = f"HTTP/1.1 {status.value} {status.phrase}\r\n"
-    hdr = ""
-    if headers:
-        for k, v in headers.items():
-            hdr += f"{k}: {v}\r\n"
-    response = (line + hdr + "\r\n").encode()
-    if not head_only:
-        response += body if isinstance(body, bytes) else body.encode()
-    writer.write(response)
-
-
-# ---------------------------------------------------------------------------
-# WebSocket — broadcast gateway
-# ---------------------------------------------------------------------------
-ws_clients: Set[websockets.asyncio.server.ServerConnection] = set()
-
-
-async def ws_handler(websocket: websockets.asyncio.server.ServerConnection):
-    ws_clients.add(websocket)
-    log.info("WS client connected from %s. Total: %d", websocket.remote_address, len(ws_clients))
+async def broadcast_handler(websocket: websockets.WebSocketServerProtocol):
+    """Handles individual client connections and message broadcasting."""
+    clients.add(websocket)
+    addr = websocket.remote_address
+    logger.info(f"Client connected from {addr}. Total clients: {len(clients)}")
+    
     try:
         async for message in websocket:
+            # Parse for logging/validation if it's JSON
             try:
                 data = json.loads(message)
                 msg_type = data.get("type", "unknown")
-                if msg_type in ("agent_register", "thought", "action"):
-                    log.debug("WS %s from %s", msg_type, websocket.remote_address)
+                # Optional: log specific important message types
+                if msg_type in ["agent_register", "thought", "action"]:
+                    logger.debug(f"Received {msg_type} from {addr}")
             except (json.JSONDecodeError, TypeError):
                 pass
 
             # Broadcast to all OTHER clients
+            if not clients:
+                continue
+                
             disconnected = set()
-            for client in ws_clients:
+            # Create broadcast tasks, tracking which client each task targets
+            task_client_pairs = []
+            for client in clients:
                 if client != websocket and client.open:
-                    try:
-                        await client.send(message)
-                    except Exception:
-                        disconnected.add(client)
-            ws_clients.difference_update(disconnected)
+                    task = asyncio.create_task(client.send(message))
+                    task_client_pairs.append((task, client))
 
+            if task_client_pairs:
+                tasks = [pair[0] for pair in task_client_pairs]
+                results = await asyncio.gather(*tasks, return_exceptions=True)
+                for i, result in enumerate(results):
+                    if isinstance(result, Exception):
+                        target_client = task_client_pairs[i][1]
+                        logger.error(f"Failed to send to client {target_client.remote_address}: {result}")
+                        disconnected.add(target_client)
+            
+            if disconnected:
+                clients.difference_update(disconnected)
+                
     except websockets.exceptions.ConnectionClosed:
-        pass
-    except Exception as exc:
-        log.error("WS handler error for %s: %s", websocket.remote_address, exc)
+        logger.debug(f"Connection closed by client {addr}")
+    except Exception as e:
+        logger.error(f"Error handling client {addr}: {e}")
     finally:
-        ws_clients.discard(websocket)
-        log.info("WS client disconnected. Total: %d", len(ws_clients))
+        clients.discard(websocket)
+        logger.info(f"Client disconnected {addr}. Total clients: {len(clients)}")
 
-
-# ---------------------------------------------------------------------------
-# Main — run both servers concurrently
-# ---------------------------------------------------------------------------
 async def main():
-    # HTTP server
-    http_server = await asyncio.start_server(http_handler, HOST, HTTP_PORT)
-    log.info("HTTP server listening on http://%s:%d", HOST, HTTP_PORT)
-
-    # WebSocket server
-    ws_server = await websockets.asyncio.server.serve(ws_handler, HOST, WS_PORT)
-    log.info("WebSocket server listening on ws://%s:%d", HOST, WS_PORT)
-    log.info("Nexus is live — open http://%s:%d in a browser", HOST, HTTP_PORT)
-
-    # Graceful shutdown
+    """Main server loop with graceful shutdown."""
+    logger.info(f"Starting Nexus WS gateway on ws://{HOST}:{PORT}")
+    
+    # Set up signal handlers for graceful shutdown
     loop = asyncio.get_running_loop()
     stop = loop.create_future()
-
+    
     def shutdown():
         if not stop.done():
             stop.set_result(None)
@@ -193,28 +95,29 @@ async def main():
         try:
             loop.add_signal_handler(sig, shutdown)
         except NotImplementedError:
+            # Signal handlers not supported on Windows
             pass
 
-    await stop
-    log.info("Shutting down...")
-
-    http_server.close()
-    await http_server.wait_closed()
-    ws_server.close()
-    await ws_server.wait_closed()
-
-    remaining = {c for c in ws_clients if c.open}
+    async with websockets.serve(broadcast_handler, HOST, PORT):
+        logger.info("Gateway is ready and listening.")
+        await stop
+        
+    logger.info("Shutting down Nexus WS gateway...")
+    # Close any remaining client connections (handlers may have already cleaned up)
+    remaining = {c for c in clients if c.open}
     if remaining:
-        await asyncio.gather(*(c.close() for c in remaining), return_exceptions=True)
-    ws_clients.clear()
-    log.info("Shutdown complete.")
-
+        logger.info(f"Closing {len(remaining)} active connections...")
+        close_tasks = [client.close() for client in remaining]
+        await asyncio.gather(*close_tasks, return_exceptions=True)
+    clients.clear()
+    
+    logger.info("Shutdown complete.")
 
 if __name__ == "__main__":
     try:
         asyncio.run(main())
     except KeyboardInterrupt:
         pass
-    except Exception as exc:
-        log.critical("Fatal: %s", exc)
+    except Exception as e:
+        logger.critical(f"Fatal server error: {e}")
         sys.exit(1)

tests/test_agent_memory.py

@@ -0,0 +1,377 @@
+"""
+Tests for agent memory — cross-session agent memory via MemPalace.
+
+Tests the memory module, hooks, and session mining without requiring
+a live ChromaDB instance. Uses mocking for the MemPalace backend.
+"""
+
+from __future__ import annotations
+
+import json
+import tempfile
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from agent.memory import (
+    AgentMemory,
+    MemoryContext,
+    SessionTranscript,
+    create_agent_memory,
+)
+from agent.memory_hooks import MemoryHooks
+
+
+# ---------------------------------------------------------------------------
+# SessionTranscript tests
+# ---------------------------------------------------------------------------
+
+
+class TestSessionTranscript:
+    def test_create(self):
+        t = SessionTranscript(agent_name="test", wing="wing_test")
+        assert t.agent_name == "test"
+        assert t.wing == "wing_test"
+        assert len(t.entries) == 0
+
+    def test_add_user_turn(self):
+        t = SessionTranscript(agent_name="test", wing="wing_test")
+        t.add_user_turn("Hello")
+        assert len(t.entries) == 1
+        assert t.entries[0]["role"] == "user"
+        assert t.entries[0]["text"] == "Hello"
+
+    def test_add_agent_turn(self):
+        t = SessionTranscript(agent_name="test", wing="wing_test")
+        t.add_agent_turn("Response")
+        assert t.entries[0]["role"] == "agent"
+
+    def test_add_tool_call(self):
+        t = SessionTranscript(agent_name="test", wing="wing_test")
+        t.add_tool_call("shell", "ls", "file1 file2")
+        assert t.entries[0]["role"] == "tool"
+        assert t.entries[0]["tool"] == "shell"
+
+    def test_summary_empty(self):
+        t = SessionTranscript(agent_name="test", wing="wing_test")
+        assert t.summary() == "Empty session."
+
+    def test_summary_with_entries(self):
+        t = SessionTranscript(agent_name="test", wing="wing_test")
+        t.add_user_turn("Do something")
+        t.add_agent_turn("Done")
+        t.add_tool_call("shell", "ls", "ok")
+
+        summary = t.summary()
+        assert "USER: Do something" in summary
+        assert "AGENT: Done" in summary
+        assert "TOOL(shell): ok" in summary
+
+    def test_text_truncation(self):
+        t = SessionTranscript(agent_name="test", wing="wing_test")
+        long_text = "x" * 5000
+        t.add_user_turn(long_text)
+        assert len(t.entries[0]["text"]) == 2000
+
+
+# ---------------------------------------------------------------------------
+# MemoryContext tests
+# ---------------------------------------------------------------------------
+
+
+class TestMemoryContext:
+    def test_empty_context(self):
+        ctx = MemoryContext()
+        assert ctx.to_prompt_block() == ""
+
+    def test_unloaded_context(self):
+        ctx = MemoryContext()
+        ctx.loaded = False
+        assert ctx.to_prompt_block() == ""
+
+    def test_loaded_with_data(self):
+        ctx = MemoryContext()
+        ctx.loaded = True
+        ctx.recent_diaries = [
+            {"text": "Fixed PR #1386", "timestamp": "2026-04-13T10:00:00Z"}
+        ]
+        ctx.facts = [
+            {"text": "Bezalel runs on VPS Beta", "score": 0.95}
+        ]
+        ctx.relevant_memories = [
+            {"text": "Changed CI runner", "score": 0.87}
+        ]
+
+        block = ctx.to_prompt_block()
+        assert "Recent Session Summaries" in block
+        assert "Fixed PR #1386" in block
+        assert "Known Facts" in block
+        assert "Bezalel runs on VPS Beta" in block
+        assert "Relevant Past Memories" in block
+
+    def test_loaded_empty(self):
+        ctx = MemoryContext()
+        ctx.loaded = True
+        # No data — should return empty string
+        assert ctx.to_prompt_block() == ""
+
+
+# ---------------------------------------------------------------------------
+# AgentMemory tests (with mocked MemPalace)
+# ---------------------------------------------------------------------------
+
+
+class TestAgentMemory:
+    def test_create(self):
+        mem = AgentMemory(agent_name="bezalel")
+        assert mem.agent_name == "bezalel"
+        assert mem.wing == "wing_bezalel"
+
+    def test_custom_wing(self):
+        mem = AgentMemory(agent_name="bezalel", wing="custom_wing")
+        assert mem.wing == "custom_wing"
+
+    def test_factory(self):
+        mem = create_agent_memory("ezra")
+        assert mem.agent_name == "ezra"
+        assert mem.wing == "wing_ezra"
+
+    def test_unavailable_graceful(self):
+        """Test graceful degradation when MemPalace is unavailable."""
+        mem = AgentMemory(agent_name="test")
+        mem._available = False  # Force unavailable
+
+        # Should not raise
+        ctx = mem.recall_context("test query")
+        assert ctx.loaded is False
+        assert ctx.error == "MemPalace unavailable"
+
+        # remember returns None
+        assert mem.remember("test") is None
+
+        # search returns empty
+        assert mem.search("test") == []
+
+    def test_start_end_session(self):
+        mem = AgentMemory(agent_name="test")
+        mem._available = False
+
+        transcript = mem.start_session()
+        assert isinstance(transcript, SessionTranscript)
+        assert mem._transcript is not None
+
+        doc_id = mem.end_session()
+        assert mem._transcript is None
+
+    def test_remember_graceful_when_unavailable(self):
+        """Test remember returns None when MemPalace is unavailable."""
+        mem = AgentMemory(agent_name="test")
+        mem._available = False
+
+        doc_id = mem.remember("some important fact")
+        assert doc_id is None
+
+    def test_write_diary_from_transcript(self):
+        mem = AgentMemory(agent_name="test")
+        mem._available = False
+
+        transcript = mem.start_session()
+        transcript.add_user_turn("Hello")
+        transcript.add_agent_turn("Hi there")
+
+        # Write diary should handle unavailable gracefully
+        doc_id = mem.write_diary()
+        assert doc_id is None  # MemPalace unavailable
+
+
+# ---------------------------------------------------------------------------
+# MemoryHooks tests
+# ---------------------------------------------------------------------------
+
+
+class TestMemoryHooks:
+    def test_create(self):
+        hooks = MemoryHooks(agent_name="bezalel")
+        assert hooks.agent_name == "bezalel"
+        assert hooks.is_active is False
+
+    def test_session_lifecycle(self):
+        hooks = MemoryHooks(agent_name="test")
+
+        # Force memory unavailable
+        hooks._memory = AgentMemory(agent_name="test")
+        hooks._memory._available = False
+
+        # Start session
+        block = hooks.on_session_start()
+        assert hooks.is_active is True
+        assert block == ""  # No memory available
+
+        # Record turns
+        hooks.on_user_turn("Hello")
+        hooks.on_agent_turn("Hi")
+        hooks.on_tool_call("shell", "ls", "ok")
+
+        # Record decision
+        hooks.on_important_decision("Switched to self-hosted CI")
+
+        # End session
+        doc_id = hooks.on_session_end()
+        assert hooks.is_active is False
+
+    def test_hooks_before_session(self):
+        """Hooks before session start should be no-ops."""
+        hooks = MemoryHooks(agent_name="test")
+        hooks._memory = AgentMemory(agent_name="test")
+        hooks._memory._available = False
+
+        # Should not raise
+        hooks.on_user_turn("Hello")
+        hooks.on_agent_turn("Response")
+
+    def test_hooks_after_session_end(self):
+        """Hooks after session end should be no-ops."""
+        hooks = MemoryHooks(agent_name="test")
+        hooks._memory = AgentMemory(agent_name="test")
+        hooks._memory._available = False
+
+        hooks.on_session_start()
+        hooks.on_session_end()
+
+        # Should not raise
+        hooks.on_user_turn("Late message")
+        doc_id = hooks.on_session_end()
+        assert doc_id is None
+
+    def test_search_during_session(self):
+        hooks = MemoryHooks(agent_name="test")
+        hooks._memory = AgentMemory(agent_name="test")
+        hooks._memory._available = False
+
+        results = hooks.search("some query")
+        assert results == []
+
+
+# ---------------------------------------------------------------------------
+# Session mining tests
+# ---------------------------------------------------------------------------
+
+
+class TestSessionMining:
+    def test_parse_session_file(self):
+        from bin.memory_mine import parse_session_file
+
+        with tempfile.NamedTemporaryFile(mode="w", suffix=".jsonl", delete=False) as f:
+            f.write('{"role": "user", "content": "Hello"}\n')
+            f.write('{"role": "assistant", "content": "Hi there"}\n')
+            f.write('{"role": "tool", "name": "shell", "content": "ls output"}\n')
+            f.write("\n")  # blank line
+            f.write("not json\n")  # malformed
+            path = Path(f.name)
+
+        turns = parse_session_file(path)
+        assert len(turns) == 3
+        assert turns[0]["role"] == "user"
+        assert turns[1]["role"] == "assistant"
+        assert turns[2]["role"] == "tool"
+        path.unlink()
+
+    def test_summarize_session(self):
+        from bin.memory_mine import summarize_session
+
+        turns = [
+            {"role": "user", "content": "Check CI"},
+            {"role": "assistant", "content": "Running CI check..."},
+            {"role": "tool", "name": "shell", "content": "5 tests passed"},
+            {"role": "assistant", "content": "CI is healthy"},
+        ]
+
+        summary = summarize_session(turns, "bezalel")
+        assert "bezalel" in summary
+        assert "Check CI" in summary
+        assert "shell" in summary
+
+    def test_summarize_empty(self):
+        from bin.memory_mine import summarize_session
+
+        assert summarize_session([], "test") == "Empty session."
+
+    def test_find_session_files(self, tmp_path):
+        from bin.memory_mine import find_session_files
+
+        # Create some test files
+        (tmp_path / "session1.jsonl").write_text("{}\n")
+        (tmp_path / "session2.jsonl").write_text("{}\n")
+        (tmp_path / "notes.txt").write_text("not a session")
+
+        files = find_session_files(tmp_path, days=365)
+        assert len(files) == 2
+        assert all(f.suffix == ".jsonl" for f in files)
+
+    def test_find_session_files_missing_dir(self):
+        from bin.memory_mine import find_session_files
+
+        files = find_session_files(Path("/nonexistent/path"), days=7)
+        assert files == []
+
+    def test_mine_session_dry_run(self, tmp_path):
+        from bin.memory_mine import mine_session
+
+        session_file = tmp_path / "test.jsonl"
+        session_file.write_text(
+            '{"role": "user", "content": "Hello"}\n'
+            '{"role": "assistant", "content": "Hi"}\n'
+        )
+
+        result = mine_session(session_file, wing="wing_test", dry_run=True)
+        assert result is None  # dry run doesn't store
+
+    def test_mine_session_empty_file(self, tmp_path):
+        from bin.memory_mine import mine_session
+
+        session_file = tmp_path / "empty.jsonl"
+        session_file.write_text("")
+
+        result = mine_session(session_file, wing="wing_test")
+        assert result is None
+
+
+# ---------------------------------------------------------------------------
+# Integration test — full lifecycle
+# ---------------------------------------------------------------------------
+
+
+class TestFullLifecycle:
+    """Test the full session lifecycle without a real MemPalace backend."""
+
+    def test_full_session_flow(self):
+        hooks = MemoryHooks(agent_name="bezalel")
+
+        # Force memory unavailable
+        hooks._memory = AgentMemory(agent_name="bezalel")
+        hooks._memory._available = False
+
+        # 1. Session start
+        context_block = hooks.on_session_start("What CI issues do I have?")
+        assert isinstance(context_block, str)
+
+        # 2. User asks question
+        hooks.on_user_turn("Check CI pipeline health")
+
+        # 3. Agent uses tool
+        hooks.on_tool_call("shell", "pytest tests/", "12 passed")
+
+        # 4. Agent responds
+        hooks.on_agent_turn("CI pipeline is healthy. All 12 tests passing.")
+
+        # 5. Important decision
+        hooks.on_important_decision("Decided to keep current CI runner", room="forge")
+
+        # 6. More interaction
+        hooks.on_user_turn("Good, check memory integration next")
+        hooks.on_agent_turn("Will test agent.memory module")
+
+        # 7. Session end
+        doc_id = hooks.on_session_end()
+        assert hooks.is_active is False