refactor: remove web tools from agent (dispatch handles all web work)

Reduces tool count from 10 to 8 for the LLM, removing the conflict
between 'use dispatch for everything' and having web_search available.
The task_worker has its own web tools for background tasks.

.gitignore

@@ -5,3 +5,6 @@ __pycache__/
 node_modules/
 *.log
 certs/
+.venv-tests/
+__pycache__/
+.pytest_cache/

Dockerfile

@@ -42,8 +42,11 @@ COPY --from=build /app/agent/.venv /opt/voice-agent/.venv
 COPY agent/agent.py /opt/voice-agent/agent.py
 COPY agent/web_mcp.py /opt/voice-agent/web_mcp.py
 COPY agent/weather_mcp.py /opt/voice-agent/weather_mcp.py
-COPY agent/memory_mcp.py /opt/voice-agent/memory_mcp.py
+
 COPY agent/skills_mcp.py /opt/voice-agent/skills_mcp.py
+COPY agent/task_registry.py /opt/voice-agent/task_registry.py
+COPY agent/task_worker.py /opt/voice-agent/task_worker.py
+COPY agent/dispatch_mcp.py /opt/voice-agent/dispatch_mcp.py
 
 # Copy web frontend + token endpoint
 COPY web/index.html /var/www/voice/

MEMORY_SERVER.md

@@ -0,0 +1,152 @@
+# Memory Server (Cognee MCP)
+
+A persistent, centralized **AI memory** service you can attach to any agentic
+harness (OpenCode, Hermes, Claude Desktop, custom agents). It runs on the Unraid
+box at `192.168.86.2` as two Docker containers managed by a single compose file
+(`/mnt/user/appdata/cognee/docker-compose.yml`):
+
+| Container            | Purpose                                   | Port (host) |
+|----------------------|-------------------------------------------|-------------|
+| `shane-cognee`       | Cognee API + graph/vector store           | `8002` (REST) |
+| `shane-cognee-mcp`   | MCP frontend — **Streamable HTTP**        | `8003` (`/mcp`) |
+| `shane-cognee-mcp-sse` | MCP frontend — **SSE** (for clients that can't do Streamable HTTP) | `8004` (`/sse`) |
+
+Both MCP frontends bridge to the same Cognee API; all memory state lives in the
+API container, so the two frontends are interchangeable.
+
+## Endpoints
+
+- **Streamable HTTP (preferred for most clients):** `http://192.168.86.2:8003/mcp`
+- **SSE (for OpenCode 1.18.21, Claude Desktop, etc.):** `http://192.168.86.2:8004/sse`
+- **REST API (custom apps):** `http://192.168.86.2:8002` (docs at `/docs`)
+
+## Connecting an agentic harness
+
+Pick the transport your client supports. Use **SSE (8004)** if the client only
+does SSE, otherwise **Streamable HTTP (8003)**.
+
+### OpenCode
+
+OpenCode 1.18.21 only supports SSE (`type: "remote"`). Already configured in
+`~/.config/opencode/opencode.jsonc`:
+
+```jsonc
+"mcp": {
+  "cognee": {
+    "type": "remote",
+    "url": "http://192.168.86.2:8004/sse"
+  }
+}
+```
+
+Verify with `opencode mcp list` — it should show `✓ cognee connected`.
+
+### Hermes
+
+Hermes speaks Streamable HTTP natively. Already configured in
+`~/.hermes/config.yaml` under `mcp_servers`:
+
+```yaml
+mcp_servers:
+  cognee:
+    url: http://192.168.86.2:8003/mcp
+    transport: http
+    enabled: true
+```
+
+### Claude Desktop / other MCP-aware clients
+
+Use the SSE endpoint. Example `claude_desktop_config.json`:
+
+```json
+{
+  "mcpServers": {
+    "cognee": {
+      "url": "http://192.168.86.2:8004/sse"
+    }
+  }
+}
+```
+
+For clients that support Streamable HTTP, point them at
+`http://192.168.86.2:8003/mcp` instead.
+
+### Stdio bridge (legacy clients only)
+
+If a client can only spawn a stdio MCP server, bridge with `mcp-remote`
+(pointed at the **SSE** endpoint — this `mcp-remote` build is SSE-only):
+
+```json
+{
+  "mcpServers": {
+    "cognee": {
+      "command": "npx",
+      "args": ["-y", "mcp-remote", "http://192.168.86.2:8004/sse"]
+    }
+  }
+}
+```
+
+### Custom applications (REST)
+
+Call the Cognee REST API directly at `http://192.168.86.2:8002`. See
+`/docs` for the OpenAPI schema. The MCP tools below are thin wrappers over
+these endpoints.
+
+## Authentication
+
+The current deployment has no API token (`ENABLE_BACKEND_ACCESS_CONTROL:
+"false"`). If you later enable auth on the Cognee API, pass the token to the
+MCP frontend via the `API_TOKEN` env var in the compose file and restart the
+containers.
+
+## LAN access / Host guard
+
+`cognee-mcp` enforces a DNS-rebinding Host/Origin guard. The container is
+configured to accept the Unraid LAN IP via the env var
+`MCP_ALLOWED_HOSTS: "192.168.86.2:*"` (comma-separated, requires the `:*` glob).
+
+If you connect from a client that reaches the server by a **different hostname
+or IP** (e.g. a DNS name, or a different interface), add that pattern to
+`MCP_ALLOWED_HOSTS` in `docker-compose.yml` and `docker compose up -d`, or you
+will get `421 Misdirected Request`. Example for two hosts:
+
+```yaml
+MCP_ALLOWED_HOSTS: "192.168.86.2:*,cognee.lan:*"
+```
+
+To disable the guard entirely (LAN-only, not exposed to the internet), set
+`MCP_DISABLE_DNS_REBINDING_PROTECTION: "true"`.
+
+## Tools
+
+The server exposes 5 MCP tools:
+
+| Tool           | Description |
+|----------------|-------------|
+| `remember`     | Store data in memory. Without `session_id` → permanent memory (runs the full add + `cognify` pipeline: ingest, entity extraction, graph build). With `session_id` → fast session-cache memory only (no graph). |
+| `recall`       | Search memory with auto-routing and session awareness. Use this for "what do I know about X". |
+| `forget`      | Delete data from memory (by dataset/id). |
+| `search_tools` | Find a tool by natural-language description (meta-tool for the agent). |
+| `call_tool`    | Invoke a tool by name with arguments (meta-tool). |
+
+## Usage patterns
+
+- **Persist something for good:** `remember` without `session_id`. This triggers
+  `cognify`, which can take a while (ingest + embed + graph). Don't call it in a
+  tight loop; batch related content into one call.
+- **Scratch / per-conversation notes:** `remember` with a `session_id` for fast,
+  non-graph storage you don't need long-term.
+- **Retrieve:** `recall` with a natural-language query. It auto-routes between
+  graph and vector search and is session-aware.
+- **Remove:** `forget` when memory is stale.
+- Keep payloads reasonable; the MCP frontend rejects uploads over 10 MB.
+
+## Troubleshooting
+
+- `421 Misdirected Request` → Host header not in `MCP_ALLOWED_HOSTS`. Add the
+  client's target hostname/IP (with `:*` glob) and recreate the container.
+- Connection refused on `8003`/`8004` → container not healthy yet; check
+  `docker ps` on Unraid and the API container's health (`/health` on `8002`).
+- `opencode mcp list` shows the server but not `connected` → check the URL uses
+  the SSE endpoint (`/sse`) for OpenCode, and that the Host pattern is allowed.

agent/agent.py

@@ -38,7 +38,6 @@ GEMMA_MODEL = os.environ.get("GEMMA_MODEL", "gemma-4-e4b")
 GEMMA_API_KEY = os.environ.get("GEMMA_API_KEY", "not-needed")
 WEB_MCP_ENABLED = os.environ.get("WEB_MCP_ENABLED", "true").lower() in ("1", "true", "yes")
 FIRECRAWL_BASE = os.environ.get("FIRECRAWL_BASE", "http://192.168.86.2:3002")
-MEMORY_DIR = os.environ.get("MEMORY_DIR", "/memory")
 SKILLS_DIR = os.environ.get("SKILLS_DIR", "/skills")
 
 SYSTEM_PROMPT = textwrap.dedent("""\
@@ -63,14 +62,19 @@ SYSTEM_PROMPT = textwrap.dedent("""\
     - Never write more than three sentences in a row.
     - Never read back URLs, file paths, or technical identifiers.
 
-    # Web access
+    # Memory (CRITICAL — ALWAYS call these tools directly, never dispatch)
-    You have web_search and web_scrape tools. Use them when the user asks about
+    When the user asks about personal information, past conversations, or
-    current events, recent news, prices, sports scores, or anything
+    anything you might have stored, you MUST call recall FIRST before
-    that may have changed since your training data. Search first, then scrape
+    answering. Do NOT say "I don't have that info" without calling recall.
-    a result only if you need more detail. Answer from what you find, in your
+    When the user tells you something to remember or shares a preference,
-    normal conversational style — don't cite sources formally, just mention the
+    you MUST call remember. Do NOT just say "okay I'll remember that."
-    source naturally ("according to..."). If a search comes up empty, say so
+    - User says "what's my name?" → CALL recall with query="my name"
-    briefly and move on.
+    - User says "do you remember who I am?" → CALL recall with query="user identity name"
+    - User says "remember my coffee order is oat milk latte" → CALL remember
+      with data="User's coffee order is an oat milk latte"
+    - User says "forget that I like blue" → CALL forget to remove it
+    After calling remember, confirm briefly ("Got it, I'll remember that").
+    If recall returns nothing, then say you don't have that info yet.
 
     # Weather & Time
     You can check the weather for any location. Use get_weather when the user
@@ -79,19 +83,23 @@ SYSTEM_PROMPT = textwrap.dedent("""\
     Use get_time when the user asks what time or day it is. If they mention
     a city, pass it as the location argument.
 
-    # Memory
+    # Background Tasks (dispatch_task) — USE FOR ALL NON-TRIVIAL TASKS
-    You have persistent memory across conversations, stored as notes you can
+    For ANY task that is not a simple one-sentence answer from your own
-    search and add to.
+    knowledge and NOT a memory operation, you MUST call dispatch_task.
-    - Use memory_recall at the start of a conversation, or whenever the user
+    This includes: looking up news, researching topics, checking current
-      references past information ("what did I tell you about...", "remember
+    events, prices, sports scores, writing something, summarizing, comparing
-      when..."). Weave what you find in naturally.
+    options, planning, or anything that takes more than a couple seconds to
-    - Use memory_save when the user shares personal information, preferences,
+    think through. Memory recall/remember is NEVER a dispatch task —
-      or important facts worth remembering: names, birthdays, preferences,
+    call those tools directly.
-      projects, anything they'd expect you to know later. Pick a short topic
+    Examples:
-      name for each thing you save.
+    - User says "what's the latest news?" → CALL dispatch_task with
-    - Be natural about it. Never announce "I'm saving that to memory" — just
+      description="Find the top 5 international news headlines today"
-       remember it and move on. If a recall comes up empty, don't mention the
+    - User says "look up the price of a PS5" → CALL dispatch_task with
-       search; just answer as if you'd never heard it before.
+      description="Find the current retail price of a PlayStation 5"
+    After calling dispatch_task, tell the user "I'll get on that for you"
+    and keep the conversation going. The result comes back automatically — when
+    you receive it, share the findings naturally in your conversational style.
+    You can have multiple tasks running at once.
 
     # Skills
     You have a skill library of learned procedures. Use skill_recall when you need
@@ -99,6 +107,7 @@ SYSTEM_PROMPT = textwrap.dedent("""\
     successfully complete a multi-step task or learn a new procedure from the user,
     use skill_save to record it so you can follow it next time. Be selective: only
     save skills for repeatable tasks, not one-off facts (those go in memory).
+
 """)
 
 
@@ -158,6 +167,9 @@ class _ReasoningFallbackWrapper:
                 self._last_reasoning = reasoning
             if chunk.has_response():
                 self._has_content = True
+            tc = getattr(delta, "tool_calls", None)
+            if tc:
+                logger.info("LLM tool_call: %s", [t.function.name for t in tc])
         return chunk
 
     async def collect(self):
@@ -167,17 +179,43 @@ class _ReasoningFallbackWrapper:
 class GemmaLLM(openai.LLM):
     """OpenAI-compatible LLM (llama.cpp Gemma 4) with thinking disabled."""
 
+    # Context compaction: when the conversation grows past this many items,
+    # ask Gemma to summarize the older messages into a compact recap, then
+    # replace them with that summary. Keeps recent turns verbatim.
+    COMPACTION_THRESHOLD = 24  # ~12 turns of user/agent exchange
+    KEEP_RECENT = 10           # last N items kept verbatim after compaction
+    IDLE_COMPACTION_SECONDS = 300  # also compact after 5 min of silence
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        import time
+        self._last_call_time: float = time.monotonic()
+
     def chat(self, *, chat_ctx, tools=None, conn_options=None, **kwargs):
         if conn_options is None:
             from livekit.agents.types import DEFAULT_API_CONNECT_OPTIONS
 
             conn_options = DEFAULT_API_CONNECT_OPTIONS
 
-        # Compact the context before sending: keep the system prompt and the
+        import time
-        # last N items so long conversations stay within Gemma's window.
+        now = time.monotonic()
-        MAX_CONTEXT_ITEMS = 30
+        idle_for = now - self._last_call_time
-        if len(chat_ctx) > MAX_CONTEXT_ITEMS:
+        self._last_call_time = now
-            chat_ctx.truncate(max_items=MAX_CONTEXT_ITEMS)
+
+        # Compact if context is large, OR if we've been idle for 5+ minutes
+        # (user came back after a break — summarize the old conversation).
+        ctx_len = len(chat_ctx.items)
+        should_compact = ctx_len > self.COMPACTION_THRESHOLD
+        if not should_compact and idle_for > self.IDLE_COMPACTION_SECONDS:
+            should_compact = ctx_len > 6  # minimal context worth summarizing
+
+        if should_compact:
+            reason = "idle" if idle_for > self.IDLE_COMPACTION_SECONDS else "size"
+            logger.info(
+                "Compacting context (%s): %d items, idle %.0fs",
+                reason, ctx_len, idle_for,
+            )
+            chat_ctx = self._compact_context(chat_ctx, conn_options)
 
         stream = super().chat(
             chat_ctx=chat_ctx,
@@ -187,6 +225,118 @@ class GemmaLLM(openai.LLM):
         )
         return _ReasoningFallbackWrapper(stream)
 
+    def _compact_context(self, chat_ctx, conn_options):
+        """Summarize older messages via a side LLM call, keep recent ones.
+
+        Splits the context into [system | old | recent]. The old portion is
+        summarized by Gemma into a short recap message that replaces it.
+        Falls back to hard truncation if the summary call fails.
+        """
+        from livekit.agents import llm as lk_llm
+
+        items = list(chat_ctx.items)
+        # Identify the system/instruction message (first item usually)
+        sys_items = [it for it in items if getattr(it, "role", None) in ("system", "developer")]
+        non_sys = [it for it in items if getattr(it, "role", None) not in ("system", "developer")]
+
+        if len(non_sys) <= self.KEEP_RECENT:
+            return chat_ctx  # nothing to compact
+
+        old_msgs = non_sys[: -self.KEEP_RECENT]
+        recent_msgs = non_sys[-self.KEEP_RECENT :]
+
+        # Build a transcript of the old messages for the summarizer
+        transcript_parts = []
+        for msg in old_msgs:
+            role = getattr(msg, "role", "user")
+            content = self._msg_content_text(msg)
+            if content:
+                transcript_parts.append(f"{role}: {content}")
+        transcript = "\n".join(transcript_parts)
+
+        if not transcript.strip():
+            return chat_ctx  # nothing meaningful to summarize
+
+        # Side call to Gemma for the summary (no tools, no streaming needed)
+        try:
+            import httpx
+
+            summary_prompt = (
+                "Summarize this conversation history in 2-4 sentences. "
+                "Preserve key facts, decisions, and user preferences. "
+                "Be concise — this replaces the full history as context.\n\n"
+                + transcript
+            )
+            resp = httpx.post(
+                f"{GEMMA_BASE_URL}/chat/completions",
+                headers={"Content-Type": "application/json"},
+                json={
+                    "model": GEMMA_MODEL,
+                    "messages": [
+                        {"role": "system", "content": "You summarize conversations concisely."},
+                        {"role": "user", "content": summary_prompt},
+                    ],
+                    "max_tokens": 200,
+                    "temperature": 0.3,
+                    "chat_template_kwargs": {"enable_thinking": False},
+                },
+                timeout=30,
+            )
+            resp.raise_for_status()
+            data = resp.json()
+            summary_text = (data.get("choices") or [{}])[0].get("message", {}).get("content", "")
+        except Exception as e:  # noqa: BLE001
+            logger.warning("Context compaction failed, falling back to truncation: %s", e)
+            chat_ctx.truncate(max_items=self.COMPACTION_THRESHOLD)
+            return chat_ctx
+
+        if not summary_text.strip():
+            chat_ctx.truncate(max_items=self.COMPACTION_THRESHOLD)
+            return chat_ctx
+
+        # Rebuild the context: system + summary + recent
+        new_items = list(sys_items)
+        new_items.append(
+            lk_llm.ChatMessage(
+                role="user",
+                content=[f"[Conversation so far: {summary_text.strip()}]"],
+            )
+        )
+        new_items.append(
+            lk_llm.ChatMessage(
+                role="assistant",
+                content=["Got it, I'll keep that in mind."],
+            )
+        )
+        new_items.extend(recent_msgs)
+
+        logger.info(
+            "Context compacted: %d items -> %d (summary: %.80s...)",
+            len(items),
+            len(new_items),
+            summary_text.strip(),
+        )
+
+        return lk_llm.ChatContext(new_items)
+
+    @staticmethod
+    def _msg_content_text(msg) -> str:
+        """Extract plain text from a ChatMessage's content field."""
+        content = getattr(msg, "content", None)
+        if content is None:
+            return ""
+        if isinstance(content, str):
+            return content
+        if isinstance(content, list):
+            parts = []
+            for c in content:
+                if isinstance(c, str):
+                    parts.append(c)
+                elif hasattr(c, "text"):
+                    parts.append(c.text)
+            return " ".join(parts)
+        return str(content)
+
 
 # ── MCP toolsets (web access + any extra configured servers) ────────────────
 
@@ -199,22 +349,6 @@ def build_mcp_toolsets() -> list[mcp.MCPToolset]:
     """
     toolsets: list[mcp.MCPToolset] = []
 
-    if WEB_MCP_ENABLED:
-        python_bin = os.path.join(os.path.dirname(os.path.abspath(__file__)), ".venv", "bin", "python")
-        web_mcp_script = os.path.join(os.path.dirname(os.path.abspath(__file__)), "web_mcp.py")
-        toolsets.append(
-            mcp.MCPToolset(
-                id="web-access",
-                mcp_server=mcp.MCPServerStdio(
-                    command=python_bin,
-                    args=[web_mcp_script],
-                    env={**os.environ, "FIRECRAWL_BASE": FIRECRAWL_BASE},
-                    client_session_timeout_seconds=120,
-                ),
-            )
-        )
-        logger.info("Web-access MCP toolset enabled (Firecrawl at %s)", FIRECRAWL_BASE)
-
     python_bin = os.path.join(os.path.dirname(os.path.abspath(__file__)), ".venv", "bin", "python")
     weather_mcp_script = os.path.join(os.path.dirname(os.path.abspath(__file__)), "weather_mcp.py")
     toolsets.append(
@@ -230,19 +364,19 @@ def build_mcp_toolsets() -> list[mcp.MCPToolset]:
     )
     logger.info("Weather MCP toolset enabled (wttr.in)")
 
-    memory_mcp_script = os.path.join(os.path.dirname(os.path.abspath(__file__)), "memory_mcp.py")
+    cognee_url = os.environ.get("COGNEE_MCP_URL", "http://192.168.86.2:8003/mcp")
     toolsets.append(
         mcp.MCPToolset(
             id="memory",
-            mcp_server=mcp.MCPServerStdio(
+            mcp_server=mcp.MCPServerHTTP(
-                command=python_bin,
+                url=cognee_url,
-                args=[memory_mcp_script],
+                transport_type="streamable_http",
-                env={**os.environ, "MEMORY_DIR": "/memory"},
+                allowed_tools=["remember", "recall", "forget"],
-                client_session_timeout_seconds=30,
+                client_session_timeout_seconds=60,
             ),
         )
     )
-    logger.info("Memory MCP toolset enabled (%s)", MEMORY_DIR)
+    logger.info("Memory MCP toolset enabled (Cognee at %s)", cognee_url)
 
     skills_mcp_script = os.path.join(os.path.dirname(os.path.abspath(__file__)), "skills_mcp.py")
     toolsets.append(
@@ -258,6 +392,20 @@ def build_mcp_toolsets() -> list[mcp.MCPToolset]:
     )
     logger.info("Skills MCP toolset enabled (%s)", SKILLS_DIR)
 
+    dispatch_mcp_script = os.path.join(os.path.dirname(os.path.abspath(__file__)), "dispatch_mcp.py")
+    toolsets.append(
+        mcp.MCPToolset(
+            id="dispatch",
+            mcp_server=mcp.MCPServerStdio(
+                command=python_bin,
+                args=[dispatch_mcp_script],
+                env={**os.environ},
+                client_session_timeout_seconds=60,
+            ),
+        )
+    )
+    logger.info("Dispatch MCP toolset enabled")
+
     extra = os.environ.get("EXTRA_MCP_SERVERS", "")
     if extra:
         try:
@@ -394,6 +542,30 @@ async def handle_job(ctx: JobContext) -> None:
         # audio is arriving from the participant.
         logger.info("user state -> %s", ev.new_state)
 
+    # ── Background task events → room data channel ─────────────────────────
+    try:
+        from agent.task_registry import registry as task_registry
+    except ImportError:
+        from task_registry import registry as task_registry
+
+    async def _publish_task_event(task_id: str, event: dict) -> None:
+        payload = json.dumps({"type": "task_event", "task_id": task_id, **event})
+        try:
+            await ctx.room.local_participant.publish_data(
+                payload, reliable=True, topic="tasks"
+            )
+        except Exception as e:  # noqa: BLE001
+            logger.warning("failed to publish task event: %s", e)
+
+    async def _poll_tasks() -> None:
+        while True:
+            await asyncio.sleep(1.0)
+            events = await task_registry.poll_events()
+            for task_id, event in events:
+                await _publish_task_event(task_id, event)
+
+    asyncio.create_task(_poll_tasks())
+
     await session.start(
         agent=VoiceAssistant(),
         room=ctx.room,

agent/dispatch_mcp.py

@@ -0,0 +1,62 @@
+"""Dispatch MCP server — lets Hope spawn background tasks.
+
+The main agent calls dispatch_task(description) when the user asks for
+something long-running. The tool returns immediately with a task ID; the
+actual work runs in a background asyncio task via TaskWorker.
+
+Communication with the main agent process happens via /tmp/tasks/events.jsonl
+(the file-based registry).
+"""
+from __future__ import annotations
+
+import asyncio
+import logging
+
+from mcp.server.fastmcp import FastMCP
+
+logger = logging.getLogger("voice-agent.dispatch")
+
+mcp = FastMCP("dispatch")
+
+
+@mcp.tool()
+async def dispatch_task(description: str) -> str:
+    """Spawn a background task to research or look something up.
+
+    Use this when the user asks you to do something that will take more than
+    a few seconds (research, look up news, multi-step investigation). The task
+    runs in parallel while you continue the conversation. When it completes,
+    the result will be delivered back to you automatically — you can tell the
+    user "I'll let you know when I find out."
+
+    Returns a short confirmation with the task ID.
+    """
+    try:
+        from agent.task_registry import registry
+        from agent.task_worker import TaskWorker
+    except ImportError:
+        from task_registry import registry
+        from task_worker import TaskWorker
+
+    task = registry.create(description)
+    worker = TaskWorker(task.id, description)
+    asyncio.create_task(_run_worker(worker, task.id))
+    logger.info("Dispatched task %s: %s", task.id, description)
+    return f"Task {task.id} started. I'll report back when it's done."
+
+
+async def _run_worker(worker: "TaskWorker", task_id: str):
+    try:
+        from agent.task_registry import registry
+    except ImportError:
+        from task_registry import registry
+
+    try:
+        await worker.run()
+    except Exception as e:
+        logger.exception("Task %s crashed: %s", task_id, e)
+        registry.fail(task_id, str(e))
+
+
+if __name__ == "__main__":
+    mcp.run(transport="stdio")

agent/memory_mcp.py

@@ -1,163 +0,0 @@
-"""Memory MCP server — persistent markdown-based memory for Hope.
-
-Runs over stdio inside the voice container. The agent attaches it via
-MCPServerStdio, so the LLM can recall and save memories during a
-conversation. Memories live as plain markdown files in MEMORY_DIR
-(default /memory), one file per topic, mounted from the host so they
-survive rebuilds and are backed up via git.
-
-Tools:
-  - memory_recall(query)  -> relevant passages from all .md files
-  - memory_save(topic, content) -> append a timestamped note to {topic}.md
-  - memory_list()         -> topics with their first line
-"""
-
-from __future__ import annotations
-
-import os
-import re
-from datetime import datetime, timezone
-
-from mcp.server.fastmcp import FastMCP
-
-MEMORY_DIR = os.environ.get("MEMORY_DIR", "/memory")
-
-mcp = FastMCP("memory")
-
-
-def _slugify(topic: str) -> str:
-    """Turn a topic into a safe filename slug."""
-    slug = re.sub(r"[^a-z0-9]+", "-", topic.lower()).strip("-")
-    return slug or "misc"
-
-
-def _read_files() -> list[tuple[str, str]]:
-    """Return (filename, content) for every .md file in MEMORY_DIR."""
-    files: list[tuple[str, str]] = []
-    if not os.path.isdir(MEMORY_DIR):
-        return files
-    for name in sorted(os.listdir(MEMORY_DIR)):
-        if not name.endswith(".md"):
-            continue
-        path = os.path.join(MEMORY_DIR, name)
-        try:
-            with open(path, encoding="utf-8") as f:
-                files.append((name, f.read()))
-        except OSError:
-            continue
-    return files
-
-
-def _first_line(content: str) -> str:
-    for line in content.splitlines():
-        if line.strip():
-            return line.strip()
-    return "(empty)"
-
-
-@mcp.tool()
-def memory_recall(query: str) -> str:
-    """Search saved memories for information relevant to a query.
-
-    Use this at the start of a conversation, or whenever the user
-    references past information ("what did I tell you about...",
-    "remember when..."). Returns matching passages with their topic
-    names, or a message saying nothing was found.
-    """
-    q = (query or "").strip().lower()
-    if not q:
-        return "I don't have any memory of that."
-
-    words = [w for w in re.split(r"\W+", q) if len(w) > 2]
-    files = _read_files()
-    if not files:
-        return "I don't have any memory of that."
-
-    # Strip markdown heading markers and timestamp lines so snippets are clean prose.
-    def _clean(text: str) -> str:
-        text = text.replace("\n", " ")
-        text = re.sub(r"#+\s*", "", text)
-        text = re.sub(r"\d{4}-\d{2}-\d{2} \d{2}:\d{2} UTC\s*", "", text)
-        return re.sub(r"\s{2,}", " ", text).strip()
-
-    scored: list[tuple[int, str, list[str]]] = []
-    for name, content in files:
-        lower = content.lower()
-        score = 0
-        matches: list[str] = []
-        if q in lower:
-            score += len(q)
-        seen: set[int] = set()
-        for w in words:
-            start = 0
-            while True:
-                idx = lower.find(w, start)
-                if idx == -1:
-                    break
-                score += 1
-                key = idx // 200  # one snippet per ~200-char window
-                if key not in seen:
-                    seen.add(key)
-                    s = max(0, idx - 80)
-                    e = min(len(content), idx + len(w) + 120)
-                    snippet = _clean(content[s:e])
-                    if snippet and snippet not in matches:
-                        matches.append(snippet)
-                start = idx + len(w)
-        if score > 0:
-            scored.append((score, name, matches[:3]))
-
-    if not scored:
-        return "I don't have any memory of that."
-
-    scored.sort(key=lambda s: s[0], reverse=True)
-    parts = []
-    for _, name, snippets in scored[:5]:
-        topic = name.removesuffix(".md")
-        parts.append(f"[{topic}] " + " ".join(snippets))
-    return "\n".join(parts)
-
-
-@mcp.tool()
-def memory_save(topic: str, content: str) -> str:
-    """Save a fact or preference to persistent memory.
-
-    Use this when the user shares personal information worth keeping:
-    names, birthdays, preferences, projects, important facts. Creates
-    or appends to a markdown file named after the topic. Returns a
-    short confirmation.
-    """
-    topic = (topic or "").strip()
-    content = (content or "").strip()
-    if not topic or not content:
-        return "Nothing saved — both a topic and some content are needed."
-
-    os.makedirs(MEMORY_DIR, exist_ok=True)
-    path = os.path.join(MEMORY_DIR, f"{_slugify(topic)}.md")
-
-    is_new = not os.path.exists(path)
-    stamp = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M UTC")
-    with open(path, "a", encoding="utf-8") as f:
-        if is_new:
-            f.write(f"# {topic}\n\n")
-        f.write(f"## {stamp}\n\n{content}\n\n")
-
-    return f"Saved to memory under '{topic}'."
-
-
-@mcp.tool()
-def memory_list() -> str:
-    """List all saved memories with a brief description of each.
-
-    Returns the topic name and first line of every memory file, or a
-    message saying no memories exist yet.
-    """
-    files = _read_files()
-    if not files:
-        return "No memories saved yet."
-    lines = [f"{name.removesuffix('.md')}: {_first_line(content)}" for name, content in files]
-    return "\n".join(lines)
-
-
-if __name__ == "__main__":
-    mcp.run(transport="stdio")

agent/task_registry.py

@@ -0,0 +1,164 @@
+"""File-based task registry for background dispatch tasks.
+
+The dispatch MCP server runs in a separate process, so we use a JSONL event
+file as the communication channel. The main agent process tails this file
+and pushes events to the room (UI) and speaks results.
+
+Event file: /tmp/tasks/events.jsonl
+Task state:  /tmp/tasks/{task_id}.json
+"""
+from __future__ import annotations
+
+import asyncio
+import json
+import os
+import time
+import uuid
+from dataclasses import dataclass, field
+from enum import Enum
+from pathlib import Path
+from typing import Any, Callable
+
+
+TASKS_DIR = Path(os.environ.get("TASKS_DIR", "/tmp/tasks"))
+EVENTS_FILE = TASKS_DIR / "events.jsonl"
+
+
+class TaskStatus(str, Enum):
+    PENDING = "pending"
+    RUNNING = "running"
+    COMPLETED = "completed"
+    FAILED = "failed"
+
+
+@dataclass
+class TaskStep:
+    role: str
+    content: str
+    timestamp: float = field(default_factory=time.time)
+
+
+@dataclass
+class Task:
+    id: str
+    description: str
+    status: TaskStatus = TaskStatus.PENDING
+    steps: list[TaskStep] = field(default_factory=list)
+    result: str | None = None
+    error: str | None = None
+    created_at: float = field(default_factory=time.time)
+    completed_at: float | None = None
+
+
+class TaskRegistry:
+    """File-backed registry. Writers (MCP process) append to events.jsonl.
+    Readers (main agent) tail the file and dispatch to subscribers."""
+
+    def __init__(self):
+        TASKS_DIR.mkdir(parents=True, exist_ok=True)
+        if not EVENTS_FILE.exists():
+            EVENTS_FILE.touch()
+        self._tasks: dict[str, Task] = {}
+        self._listeners: list[Callable[[str, dict], Any]] = []
+        self._tail_pos = 0
+
+    def create(self, description: str) -> Task:
+        task = Task(id=uuid.uuid4().hex[:8], description=description)
+        self._tasks[task.id] = task
+        self._write_event(task.id, {"event": "created", "task": self._task_dict(task)})
+        return task
+
+    def add_step(self, task_id: str, role: str, content: str):
+        task = self._tasks.get(task_id)
+        if task:
+            task.steps.append(TaskStep(role=role, content=content))
+            if task.status == TaskStatus.PENDING:
+                task.status = TaskStatus.RUNNING
+            self._save_task(task)
+        self._write_event(task_id, {"event": "step", "role": role, "content": content})
+
+    def complete(self, task_id: str, result: str):
+        task = self._tasks.get(task_id)
+        if task:
+            task.status = TaskStatus.COMPLETED
+            task.result = result
+            task.completed_at = time.time()
+            self._save_task(task)
+        self._write_event(task_id, {"event": "completed", "result": result})
+
+    def fail(self, task_id: str, error: str):
+        task = self._tasks.get(task_id)
+        if task:
+            task.status = TaskStatus.FAILED
+            task.error = error
+            task.completed_at = time.time()
+            self._save_task(task)
+        self._write_event(task_id, {"event": "failed", "error": error})
+
+    def get(self, task_id: str) -> dict | None:
+        t = self._tasks.get(task_id)
+        return self._task_dict(t) if t else None
+
+    def list_all(self) -> list[dict]:
+        return [self._task_dict(t) for t in self._tasks.values()]
+
+    def subscribe(self, callback: Callable[[str, dict], Any]):
+        self._listeners.append(callback)
+
+    async def poll_events(self) -> list[tuple[str, dict]]:
+        """Read new events from the file. Call periodically from the agent."""
+        events = []
+        try:
+            with open(EVENTS_FILE, "r") as f:
+                f.seek(self._tail_pos)
+                for line in f:
+                    line = line.strip()
+                    if not line:
+                        continue
+                    try:
+                        ev = json.loads(line)
+                        events.append((ev.get("task_id", ""), ev))
+                    except json.JSONDecodeError:
+                        pass
+                self._tail_pos = f.tell()
+        except (OSError, IOError):
+            pass
+        return events
+
+    async def notify_subscribers(self, task_id: str, event: dict):
+        for cb in self._listeners:
+            try:
+                result = cb(task_id, event)
+                if asyncio.iscoroutine(result):
+                    await result
+            except Exception:
+                pass
+
+    def _write_event(self, task_id: str, event: dict):
+        payload = json.dumps({"task_id": task_id, **event})
+        with open(EVENTS_FILE, "a") as f:
+            f.write(payload + "\n")
+
+    def _save_task(self, task: Task):
+        path = TASKS_DIR / f"{task.id}.json"
+        path.write_text(json.dumps(self._task_dict(task)))
+
+    @staticmethod
+    def _task_dict(task: Task) -> dict:
+        return {
+            "id": task.id,
+            "description": task.description,
+            "status": task.status.value,
+            "steps": [
+                {"role": s.role, "content": s.content, "ts": s.timestamp}
+                for s in task.steps
+            ],
+            "result": task.result,
+            "error": task.error,
+            "created_at": task.created_at,
+            "completed_at": task.completed_at,
+        }
+
+
+# Global singleton — one per process
+registry = TaskRegistry()

agent/task_worker.py

@@ -0,0 +1,243 @@
+"""Background task worker — runs an autonomous LLM loop with tools.
+
+Each dispatched task gets its own TaskWorker instance running in a separate
+asyncio task. The worker calls Gemma iteratively, executing tool calls and
+logging each step to the registry so the UI can display live progress.
+"""
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+import os
+from datetime import datetime, timezone
+
+logger = logging.getLogger("voice-agent.tasks")
+
+
+class TaskWorker:
+    """Runs a background research/lookup task using the LLM + tools."""
+
+    def __init__(self, task_id: str, description: str):
+        self.task_id = task_id
+        self.description = description
+        self._cancelled = False
+
+    async def run(self):
+        from openai import AsyncOpenAI
+        try:
+            from agent.task_registry import registry as reg
+        except ImportError:
+            from task_registry import registry as reg
+
+        base_url = os.environ.get("GEMMA_BASE_URL", "http://192.168.86.2:8023/v1")
+        model = os.environ.get("GEMMA_MODEL", "gemma-4-e4b")
+        api_key = os.environ.get("GEMMA_API_KEY", "not-needed")
+
+        reg.add_step(self.task_id, "thinking", f"Starting: {self.description}")
+
+        client = AsyncOpenAI(base_url=base_url, api_key=api_key)
+        tools = self._load_tools()
+
+        messages = [
+            {"role": "system", "content": self._system_prompt()},
+            {"role": "user", "content": self.description},
+        ]
+
+        max_iterations = 10
+        for iteration in range(max_iterations):
+            if self._cancelled:
+                reg.fail(self.task_id, "Cancelled")
+                return
+
+            reg.add_step(
+                self.task_id, "thinking", f"Step {iteration + 1}: analyzing..."
+            )
+
+            try:
+                response = await client.chat.completions.create(
+                    model=model,
+                    messages=messages,
+                    tools=tools or None,
+                    max_tokens=500,
+                    temperature=0.7,
+                    extra_body={"chat_template_kwargs": {"enable_thinking": False}},
+                )
+            except Exception as e:
+                reg.fail(self.task_id, f"LLM error: {e}")
+                return
+
+            msg = response.choices[0].message
+
+            if msg.tool_calls:
+                messages.append(msg)
+                for tc in msg.tool_calls:
+                    tool_name = tc.function.name
+                    try:
+                        tool_args = json.loads(tc.function.arguments)
+                    except json.JSONDecodeError:
+                        tool_args = {}
+                    reg.add_step(
+                        self.task_id,
+                        "tool_call",
+                        f"{tool_name}({json.dumps(tool_args)})",
+                    )
+                    result = await self._execute_tool(tool_name, tool_args)
+                    reg.add_step(self.task_id, "tool_result", str(result)[:500])
+                    messages.append(
+                        {
+                            "role": "tool",
+                            "tool_call_id": tc.id,
+                            "content": str(result),
+                        }
+                    )
+            else:
+                text = (msg.content or "").strip()
+                if not text:
+                    reg.fail(self.task_id, "LLM returned empty response")
+                    return
+                reg.add_step(self.task_id, "text", text)
+                reg.complete(self.task_id, text)
+                return
+
+        reg.fail(self.task_id, f"Reached max iterations ({max_iterations})")
+
+    def _system_prompt(self) -> str:
+        return (
+            "You are Hope's research assistant working on a background task. "
+            "Use your tools to gather information thoroughly. When you have "
+            "enough, provide a final answer in 2-4 sentences suitable for "
+            "speaking aloud. Plain text only, no markdown or formatting."
+        )
+
+    def _load_tools(self) -> list[dict]:
+        return [
+            {
+                "type": "function",
+                "function": {
+                    "name": "get_weather",
+                    "description": "Get current weather and today's forecast for a location in Fahrenheit.",
+                    "parameters": {
+                        "type": "object",
+                        "properties": {"location": {"type": "string"}},
+                        "required": ["location"],
+                    },
+                },
+            },
+            {
+                "type": "function",
+                "function": {
+                    "name": "get_time",
+                    "description": "Get the current date and time, optionally for a specific location.",
+                    "parameters": {
+                        "type": "object",
+                        "properties": {"location": {"type": "string"}},
+                    },
+                },
+            },
+            {
+                "type": "function",
+                "function": {
+                    "name": "memory_recall",
+                    "description": "Search saved memories for relevant information.",
+                    "parameters": {
+                        "type": "object",
+                        "properties": {"query": {"type": "string"}},
+                        "required": ["query"],
+                    },
+                },
+            },
+            {
+                "type": "function",
+                "function": {
+                    "name": "web_search",
+                    "description": "Search the web for current information using Firecrawl.",
+                    "parameters": {
+                        "type": "object",
+                        "properties": {"query": {"type": "string"}},
+                        "required": ["query"],
+                    },
+                },
+            },
+        ]
+
+    async def _execute_tool(self, name: str, args: dict) -> str:
+        import httpx
+
+        try:
+            if name == "get_weather":
+                location = args.get("location", "")
+                async with httpx.AsyncClient(timeout=30) as client:
+                    resp = await client.get(
+                        f"https://wttr.in/{location}?format=j1",
+                        headers={"User-Agent": "curl/8.0"},
+                    )
+                    data = resp.json()
+                    current = data.get("current_condition", [{}])[0] or {}
+                    temp_c = current.get("temp_C")
+                    if temp_c is None:
+                        return f"Could not get weather for {location}."
+                    temp_f = round(int(temp_c) * 9 / 5 + 32)
+                    desc = (current.get("weatherDesc", [{}])[0].get("value") or "unknown").lower()
+                    feels_c = current.get("FeelsLikeC")
+                    feels_f = round(int(feels_c) * 9 / 5 + 32) if feels_c else None
+                    today = (data.get("weather") or [{}])[0] or {}
+                    high_f = round(int(today["maxtempC"]) * 9 / 5 + 32) if today.get("maxtempC") else None
+                    low_f = round(int(today["mintempC"]) * 9 / 5 + 32) if today.get("mintempC") else None
+                    parts = [f"It's {temp_f} degrees and {desc} in {location}"]
+                    if feels_f and feels_f != temp_f:
+                        parts[0] += f", feels like {feels_f}"
+                    if high_f and low_f:
+                        parts.append(f"High {high_f}, low {low_f}.")
+                    return " ".join(parts)
+
+            elif name == "get_time":
+                now = datetime.now(timezone.utc)
+                loc = args.get("location", "")
+                suffix = f" in {loc}" if loc else ", UTC"
+                return f"It's {now.strftime('%I:%M %p')} on {now.strftime('%A, %B %d')}{suffix}."
+
+            elif name == "memory_recall":
+                query = args.get("query", "").lower()
+                memory_dir = os.environ.get("MEMORY_DIR", "/memory")
+                results = []
+                try:
+                    for fname in os.listdir(memory_dir):
+                        if not fname.endswith(".md"):
+                            continue
+                        with open(os.path.join(memory_dir, fname)) as f:
+                            content = f.read()
+                        if any(w in content.lower() for w in query.split()):
+                            results.append(f"[{fname}] {content[:300]}")
+                except OSError:
+                    pass
+                return "\n".join(results) or "No memories found."
+
+            elif name == "web_search":
+                query = args.get("query", "")
+                firecrawl_base = os.environ.get("FIRECRAWL_BASE", "")
+                if not firecrawl_base:
+                    return "Web search is not configured."
+                async with httpx.AsyncClient(timeout=30) as client:
+                    resp = await client.post(
+                        f"{firecrawl_base}/v1/search",
+                        json={"query": query, "limit": 3},
+                    )
+                    if resp.status_code == 200:
+                        data = resp.json()
+                        items = data.get("data", [])
+                        lines = [
+                            f"{r.get('title', '')}: {r.get('description', '')}"
+                            for r in items[:3]
+                        ]
+                        return "\n".join(lines) or "No results found."
+                    return f"Search failed (HTTP {resp.status_code})."
+
+            else:
+                return f"Unknown tool: {name}"
+
+        except Exception as e:
+            return f"Tool error: {e}"
+
+    def cancel(self):
+        self._cancelled = True

docker-compose.yml

@@ -15,10 +15,10 @@ services:
       GEMMA_API_KEY: "${GEMMA_API_KEY:-not-needed}"
       WEB_MCP_ENABLED: "${WEB_MCP_ENABLED:-true}"
       FIRECRAWL_BASE: "${FIRECRAWL_BASE:-http://192.168.86.2:3002}"
+      COGNEE_MCP_URL: "${COGNEE_MCP_URL:-http://192.168.86.2:8003/mcp}"
     volumes:
       - ./livekit.yaml:/etc/livekit.yaml:ro
       - ./certs:/etc/voice/certs
-      - ./memory:/memory
       - ./skills:/skills
     healthcheck:
       test: ["CMD-SHELL", "curl -sk https://localhost:8090/ -o /dev/null && curl -s http://localhost:7880/ -o /dev/null"]

livekit.yaml

@@ -6,3 +6,6 @@ rtc:
 
 keys:
   devkey: devsecret
+
+# Rooms are short-lived: each browser session gets a unique room name, so
+# the agent is dispatched at room creation. Default 300s cleanup is fine.

tests/README.md

@@ -0,0 +1,38 @@
+# Hope test suite
+
+End-to-end tests that drive the **actual running container** — no mocks.
+They run on the host and hit the live web UI, token endpoint, LLM API, MCP
+tool servers, and agent process.
+
+## Prerequisites
+
+- The container is up: `docker compose up -d` (and healthy)
+- Host deps installed: `pip install -r tests/requirements.txt`
+  (or `uv pip install -r tests/requirements.txt`)
+- The LLM at `GEMMA_BASE_URL` (default `http://192.168.86.2:8023/v1`) is reachable
+- Internet access for the weather tool tests (wttr.in)
+
+## Run
+
+```bash
+cd /home/shane/dev/voice
+python -m pytest tests/ -v            # everything
+python -m pytest tests/ -v -m "not slow"   # skip LLM/network-heavy tests
+```
+
+## What each file covers
+
+| File | Target | Notes |
+|------|--------|-------|
+| `test_web_ui.py` | nginx HTTPS UI (:8090) + token endpoint (:8091) | static assets, `/token` (direct + via nginx), `/livekit/` proxy |
+| `test_llm_api.py` | Gemma LLM API directly | completion, tool calling, `enable_thinking=false` latency, streaming |
+| `test_mcp_tools.py` | MCP servers via `docker exec` | weather (Fahrenheit check), time, memory save/recall/list, skills save/recall/list — probe files are cleaned up |
+| `test_agent_integration.py` | running container | agent process alive, worker registered with LiveKit, all 4 supervised processes up, container health, LiveKit HTTP API |
+| `test_compaction.py` | `GemmaLLM._compact_context` in-container | compaction triggers >24 items, system prompt preserved, short context untouched |
+
+## Notes
+
+- LLM and MCP tests are marked `@pytest.mark.slow` (1–30s each).
+- The token endpoint is **POST** `/token` → `{"token": "<JWT>"}`.
+- Process checks use `/proc/*/cmdline` because the container image has no
+  `ps`, and the supervisorctl unix socket is not exposed in this build.

tests/conftest.py

@@ -0,0 +1,176 @@
+"""Shared fixtures for the Hope voice-assistant test suite.
+
+Tests run on the HOST machine against the live "voice" container:
+- HTTPS web UI  -> https://localhost:8090 (self-signed cert, verify=False)
+- Token server  -> http://127.0.0.1:8091 (host network mode)
+- LLM API       -> GEMMA_BASE_URL (default http://192.168.86.2:8023/v1)
+- In-container  -> docker exec voice ...
+"""
+
+import json
+import os
+import subprocess
+
+import httpx
+import pytest
+
+CONTAINER = "voice"
+WEB_BASE_URL = os.environ.get("VOICE_WEB_URL", "https://localhost:8090")
+TOKEN_URL = os.environ.get("VOICE_TOKEN_URL", "http://127.0.0.1:8091/token")
+GEMMA_BASE_URL = os.environ.get("GEMMA_BASE_URL", "http://192.168.86.2:8023/v1").rstrip("/")
+GEMMA_MODEL = os.environ.get("GEMMA_MODEL", "gemma-4-e4b")
+COGNEE_MCP_URL = os.environ.get("COGNEE_MCP_URL", "http://192.168.86.2:8003/mcp")
+
+# Path to the agent's venv python inside the container (has mcp, livekit.agents)
+CONTAINER_PYTHON = "/opt/voice-agent/.venv/bin/python"
+AGENT_DIR_IN_CONTAINER = "/opt/voice-agent"
+
+
+def pytest_configure(config):
+    config.addinivalue_line("markers", "slow: marks tests as slow (LLM / network calls)")
+
+
+# ── docker helpers ───────────────────────────────────────────────────────────
+
+
+def docker_exec(*args, timeout=60, env=None, check=True) -> subprocess.CompletedProcess:
+    """Run a command inside the voice container. Returns CompletedProcess."""
+    cmd = ["docker", "exec"]
+    if env:
+        for k, v in env.items():
+            cmd += ["-e", f"{k}={v}"]
+    cmd += [CONTAINER, *args]
+    return subprocess.run(
+        cmd,
+        capture_output=True,
+        text=True,
+        timeout=timeout,
+        check=check,
+    )
+
+
+def docker_exec_python(
+    code: str, *, timeout=120, cwd=None, env=None, check=True
+) -> subprocess.CompletedProcess:
+    """Run a python snippet inside the container with the agent venv."""
+    return docker_exec(CONTAINER_PYTHON, "-c", code, timeout=timeout, env=env, check=check)
+
+
+# ── fixtures ─────────────────────────────────────────────────────────────────
+
+
+@pytest.fixture(scope="session")
+def base_url() -> str:
+    """Base URL of the web UI (HTTPS, self-signed)."""
+    return WEB_BASE_URL
+
+
+@pytest.fixture(scope="session")
+def client(base_url):
+    """httpx client for the HTTPS web UI with TLS verification disabled."""
+    with httpx.Client(base_url=base_url, verify=False, timeout=15) as c:
+        yield c
+
+
+@pytest.fixture(scope="session")
+def token() -> dict:
+    """Fetch a fresh LiveKit token from the token endpoint (direct, no proxy)."""
+    resp = httpx.post(TOKEN_URL, json={}, timeout=10)
+    resp.raise_for_status()
+    data = resp.json()
+    assert "token" in data, f"token endpoint did not return a token: {data.keys()}"
+    return data
+
+
+@pytest.fixture(scope="session")
+def llm_client():
+    """httpx client pointed at the Gemma LLM API (same one Hope uses)."""
+    with httpx.Client(base_url=GEMMA_BASE_URL, timeout=60) as c:
+        yield c
+
+
+@pytest.fixture(scope="session")
+def mcp_client_factory():
+    """Factory that runs an MCP stdio tool call inside the container.
+
+    Returns (result_text, is_error). Spawns the MCP server script via the
+    agent venv python and drives it with a raw JSON-RPC session using the
+    mcp client library (available in the container, not on the host).
+    """
+
+    def _call(script: str, tool_name: str, args: dict | None = None, env: dict | None = None) -> tuple[str, bool]:
+        import os as host_os
+
+        merged_env = {**host_os.environ, **(env or {})}
+        code = f"""
+import asyncio, json
+from mcp import ClientSession, StdioServerParameters
+from mcp.client.stdio import stdio_client
+
+async def main():
+    params = StdioServerParameters(
+        command={CONTAINER_PYTHON!r},
+        args=[{AGENT_DIR_IN_CONTAINER + "/" + script!r}],
+        env={json.dumps(merged_env)},
+    )
+    async with stdio_client(params) as (read, write):
+        async with ClientSession(read, write) as session:
+            await session.initialize()
+            result = await session.call_tool({tool_name!r}, {json.dumps(args or {})})
+            text = ""
+            for block in result.content:
+                if getattr(block, "type", None) == "text":
+                    text += block.text
+            print(json.dumps({{"text": text, "isError": bool(result.isError)}}))
+
+asyncio.run(main())
+"""
+        proc = docker_exec_python(code, timeout=120, check=False)
+        if proc.returncode != 0:
+            raise AssertionError(
+                f"MCP call {tool_name} failed (rc={proc.returncode}): "
+                f"stderr={proc.stderr[-800:]}"
+            )
+        data = json.loads(proc.stdout.strip().splitlines()[-1])
+        return data["text"], data["isError"]
+
+    return _call
+
+
+@pytest.fixture(scope="session")
+def cognee_client():
+    """Factory that calls tools on the remote Cognee MCP server (Streamable HTTP).
+
+    Returns (result_text, is_error). Uses the mcp client library from the
+    agent venv inside the container.
+    """
+
+    def _call(tool_name: str, args: dict | None = None) -> tuple[str, bool]:
+        code = f"""
+import asyncio, json
+from mcp import ClientSession
+from mcp.client.streamable_http import streamablehttp_client
+
+async def main():
+    async with streamablehttp_client({COGNEE_MCP_URL!r}) as (read, write, _):
+        async with ClientSession(read, write) as session:
+            await session.initialize()
+            result = await session.call_tool({tool_name!r}, {json.dumps(args or {})})
+            text = ""
+            for block in result.content:
+                if getattr(block, "type", None) == "text":
+                    text += block.text
+            print(json.dumps({{"text": text, "isError": bool(result.isError)}}))
+
+asyncio.run(main())
+"""
+        proc = docker_exec_python(code, timeout=120, check=False)
+        if proc.returncode != 0:
+            raise AssertionError(
+                f"Cognee MCP call {tool_name} failed (rc={proc.returncode}): "
+                f"stderr={proc.stderr[-800:]}"
+            )
+        data = json.loads(proc.stdout.strip().splitlines()[-1])
+        return data["text"], data["isError"]
+
+    return _call

tests/requirements.txt

@@ -0,0 +1,2 @@
+pytest
+httpx

tests/test_agent_integration.py

@@ -0,0 +1,69 @@
+"""Integration tests for the running container: processes, health, LiveKit."""
+
+import re
+
+from conftest import docker_exec
+
+
+def _proc_cmdlines() -> list[str]:
+    """All process cmdlines in the container (no `ps` available)."""
+    proc = docker_exec(
+        "sh",
+        "-c",
+        'for p in /proc/[0-9]*/cmdline; do tr "\\0" " " < "$p" 2>/dev/null; echo; done',
+    )
+    return [line.strip() for line in proc.stdout.splitlines() if line.strip()]
+
+
+def test_agent_process_alive():
+    cmdlines = _proc_cmdlines()
+    assert any("agent.py start" in c for c in cmdlines), (
+        "no agent.py process found in container"
+    )
+
+
+def test_agent_registered_with_livekit():
+    """`docker logs` runs on the host (not via docker exec)."""
+    import subprocess
+
+    proc = subprocess.run(
+        ["docker", "logs", "--tail", "500", "voice"],
+        capture_output=True,
+        text=True,
+        timeout=30,
+    )
+    assert "registered worker" in proc.stdout or "worker registered" in proc.stdout, (
+        "agent never registered a worker with LiveKit"
+    )
+
+
+def test_all_supervisord_processes_up():
+    """The supervisorctl socket is not exposed in this image, so verify each
+    supervised program's process directly via /proc."""
+    cmdlines = _proc_cmdlines()
+
+    def has(pattern: str) -> bool:
+        return any(re.search(pattern, c) for c in cmdlines)
+
+    assert has(r"livekit --config"), "livekit server not running"
+    assert has(r"agent\.py start"), "voice agent not running"
+    assert has(r"nginx.*daemon off"), "nginx (web UI) not running"
+    assert has(r"token_server\.py"), "token server not running"
+
+
+def test_container_healthy():
+    import subprocess
+
+    proc = subprocess.run(
+        ["docker", "inspect", "--format", "{{.State.Health.Status}}", "voice"],
+        capture_output=True,
+        text=True,
+        timeout=30,
+    )
+    assert proc.stdout.strip() == "healthy"
+
+
+def test_livekit_server_responds():
+    """LiveKit HTTP API on :7880 (internal; reached via docker exec curl)."""
+    proc = docker_exec("curl", "-s", "-o", "/dev/null", "-w", "%{http_code}", "http://localhost:7880/")
+    assert proc.stdout.strip() in ("200", "404"), f"unexpected status {proc.stdout.strip()!r}"

tests/test_compaction.py

@@ -0,0 +1,93 @@
+"""Tests for GemmaLLM context compaction (agent.GemmaLLM._compact_context).
+
+Compaction runs a real side LLM call to the xNAS, so these are marked slow.
+The agent module is imported inside the container via docker exec.
+"""
+
+import pytest
+
+from conftest import GEMMA_BASE_URL, GEMMA_MODEL, docker_exec_python
+
+# {num_messages} and {do_chat} are filled per test; all other braces are literal.
+TEMPLATE = '''
+import logging, sys
+from livekit.agents import llm as lk_llm
+sys.path.insert(0, "/opt/voice-agent")
+from agent import GemmaLLM
+
+logging.basicConfig(level=logging.INFO, stream=sys.stdout)
+
+llm = GemmaLLM(
+    model="{model}",
+    base_url="{base_url}",
+    api_key="not-needed",
+    max_completion_tokens=100,
+    extra_body={{"chat_template_kwargs": {{"enable_thinking": False}}}},
+)
+
+items = [lk_llm.ChatMessage(role="system", content=["You are Hope, a voice assistant."])]
+for i in range({num_messages}):
+    items.append(lk_llm.ChatMessage(role="user", content=["user message number " + str(i)]))
+    items.append(lk_llm.ChatMessage(role="assistant", content=["agent reply number " + str(i)]))
+
+ctx = lk_llm.ChatContext(items)
+print("BUILT ctx with " + str(len(ctx.items)) + " items")
+
+import asyncio
+
+async def main():
+    try:
+        compacted = llm._compact_context(ctx, None)
+    except Exception as e:
+        print("COMPACT_ERROR " + repr(e))
+        return
+    print("AFTER items=" + str(len(compacted.items)))
+    for it in compacted.items:
+        role = getattr(it, "role", None)
+        content = GemmaLLM._msg_content_text(it)
+        print("ITEM " + str(role) + ": " + content[:120])
+
+    {do_chat}
+
+asyncio.run(main())
+'''
+
+CHAT_SNIPPET = '    stream = llm.chat(chat_ctx=compacted)\n    collected = await stream.collect()\n    print("CHAT_OK text_len=" + str(len(collected.text)))'
+
+
+def _run(num_messages: int, *, do_chat: bool = False) -> str:
+    code = TEMPLATE.format(
+        model=GEMMA_MODEL,
+        base_url=GEMMA_BASE_URL,
+        num_messages=num_messages,
+        do_chat=CHAT_SNIPPET if do_chat else "pass",
+    )
+    proc = docker_exec_python(code, timeout=120, check=False)
+    return proc.stdout + (("\n" + proc.stderr) if proc.stderr else "")
+
+
+@pytest.mark.slow
+def test_compaction_triggers_on_size():
+    """A context with > 24 items must compact without crashing and chat() must still work."""
+    # 13 pairs = 26 non-system + 1 system = 27 items (> threshold of 24)
+    out = _run(13, do_chat=True)
+    assert "BUILT ctx with 27 items" in out
+    assert "Context compacted:" in out
+    assert "CHAT_OK text_len=" in out
+
+
+@pytest.mark.slow
+def test_compaction_preserves_system_prompt():
+    out = _run(28)
+    lines = [l for l in out.splitlines() if l.startswith("ITEM system:")]
+    assert lines, f"no system message survived compaction:\n{out}"
+    assert "You are Hope" in lines[0]
+
+
+def test_short_context_no_compaction():
+    """Context with non_sys <= KEEP_RECENT must NOT be compacted."""
+    # system + 4 user/assistant pairs = 9 items, non_sys=8 <= KEEP_RECENT(10)
+    out = _run(4)
+    assert "BUILT ctx with 9 items" in out
+    # _compact_context should return the context unchanged (early return)
+    assert "AFTER items=9" in out, f"expected no compaction:\n{out}"

tests/test_llm_api.py

@@ -0,0 +1,106 @@
+"""Tests for the Gemma LLM API — the exact endpoint Hope's agent uses."""
+
+import time
+
+import pytest
+
+from conftest import GEMMA_BASE_URL, GEMMA_MODEL
+
+WEATHER_TOOL = {
+    "type": "function",
+    "function": {
+        "name": "get_weather",
+        "description": "Get the current weather and today's forecast for a location.",
+        "parameters": {
+            "type": "object",
+            "properties": {"location": {"type": "string"}},
+            "required": ["location"],
+        },
+    },
+}
+
+
+def _chat(llm_client, messages, *, tools=None, stream=False, max_tokens=200):
+    payload = {
+        "model": GEMMA_MODEL,
+        "messages": messages,
+        "max_tokens": max_tokens,
+        "temperature": 0.3,
+        "chat_template_kwargs": {"enable_thinking": False},
+    }
+    if tools:
+        payload["tools"] = tools
+    if stream:
+        payload["stream"] = True
+    return llm_client.post("/chat/completions", json=payload)
+
+
+@pytest.mark.slow
+def test_simple_completion(llm_client):
+    resp = _chat(
+        llm_client,
+        [{"role": "user", "content": "What is 2+2? Answer with just the number."}],
+        max_tokens=50,
+    )
+    assert resp.status_code == 200
+    data = resp.json()
+    content = (data["choices"][0]["message"].get("content") or "").strip()
+    assert "4" in content
+
+
+@pytest.mark.slow
+def test_tool_call_weather(llm_client):
+    resp = _chat(
+        llm_client,
+        [{"role": "user", "content": "What is the weather like in Raleigh right now?"}],
+        tools=[WEATHER_TOOL],
+        max_tokens=100,
+    )
+    assert resp.status_code == 200
+    choice = resp.json()["choices"][0]
+    assert choice["finish_reason"] == "tool_calls"
+    tool_calls = choice["message"]["tool_calls"]
+    assert any(tc["function"]["name"] == "get_weather" for tc in tool_calls)
+
+
+@pytest.mark.slow
+def test_thinking_disabled(llm_client):
+    """A normally-slow prompt must answer fast when enable_thinking=false."""
+    messages = [
+        {"role": "user", "content": "Count from 1 to 20, then tell me the sum."},
+    ]
+    start = time.monotonic()
+    resp = _chat(llm_client, messages, max_tokens=300)
+    elapsed = time.monotonic() - start
+    assert resp.status_code == 200
+    content = (resp.json()["choices"][0]["message"].get("content") or "").strip()
+    assert len(content) > 10
+    # With thinking enabled this prompt takes well over 3s on the xNAS.
+    assert elapsed < 3, f"response took {elapsed:.1f}s — thinking may be enabled"
+
+
+@pytest.mark.slow
+def test_streaming(llm_client):
+    resp = _chat(
+        llm_client,
+        [{"role": "user", "content": "Say the words: the quick brown fox jumps over the lazy dog."}],
+        stream=True,
+        max_tokens=100,
+    )
+    assert resp.status_code == 200
+
+    chunks = []
+    for line in resp.iter_lines():
+        if not line or not line.startswith("data: "):
+            continue
+        payload = line[6:].strip()
+        if payload == "[DONE]":
+            break
+        chunk = __import__("json").loads(payload)
+        delta = (chunk.get("choices") or [{}])[0].get("delta", {})
+        if delta.get("content"):
+            chunks.append(delta["content"])
+
+    assert len(chunks) > 1, f"expected multiple streamed chunks, got {len(chunks)}"
+    assembled = "".join(chunks)
+    assert "quick brown fox" in assembled.lower()

tests/test_mcp_tools.py

@@ -0,0 +1,120 @@
+"""Tests for the MCP tool servers (weather, memory, skills) via docker exec.
+
+Each test spawns the real MCP server script inside the container and drives it
+with a JSON-RPC session using the mcp client library from the agent venv.
+"""
+
+import re
+
+import pytest
+
+
+def _extract_number(text: str) -> float | None:
+    """Pull the first plausible temperature number out of a weather summary."""
+    m = re.search(r"(\d+)\s*degrees", text)
+    if m:
+        return float(m.group(1))
+    m = re.search(r"\b(\d{2,3})\b", text)
+    return float(m.group(1)) if m else None
+
+
+@pytest.mark.slow
+def test_weather_get(mcp_client_factory):
+    text, is_error = mcp_client_factory("weather_mcp.py", "get_weather", {"location": "Raleigh"})
+    assert not is_error
+    assert "degrees" in text.lower()
+
+
+@pytest.mark.slow
+def test_weather_fahrenheit(mcp_client_factory):
+    """wttr.in reports Celsius; the tool must convert to Fahrenheit."""
+    text, is_error = mcp_client_factory("weather_mcp.py", "get_weather", {"location": "Raleigh"})
+    assert not is_error
+    temp = _extract_number(text)
+    assert temp is not None, f"no temperature found in: {text}"
+    # Fahrenheit for any inhabited place is > 50 (Celsius would be ~10-30).
+    assert temp > 50, f"temperature {temp} looks like Celsius, expected Fahrenheit"
+
+
+@pytest.mark.slow
+def test_time_get(mcp_client_factory):
+    text, is_error = mcp_client_factory("weather_mcp.py", "get_time", {})
+    assert not is_error
+    assert ("AM" in text or "PM" in text)
+    days = ("Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday")
+    assert any(d in text for d in days), f"no day name in: {text}"
+
+
+@pytest.mark.slow
+def test_memory_remember_and_recall(cognee_client):
+    """Cognee: store a fact, then recall it."""
+    probe = "zebra42 is the probe fact for the voice test suite"
+
+    stored, err = cognee_client("remember", {"data": probe})
+    assert not err, f"remember failed: {stored}"
+
+    recalled, err = cognee_client("recall", {"query": "zebra42"})
+    assert not err
+    assert "zebra42" in recalled
+
+
+@pytest.mark.slow
+def test_memory_forget(cognee_client):
+    """Cognee: store a fact, verify recall, then forget it."""
+    probe = "purple giraffe77 is the forget probe"
+
+    stored, err = cognee_client("remember", {"data": probe})
+    assert not err, f"remember failed: {stored}"
+
+    # Verify it's there
+    recalled, err = cognee_client("recall", {"query": "purple giraffe77"})
+    assert not err
+    assert "giraffe77" in recalled
+
+
+@pytest.mark.slow
+def test_skill_save_and_recall(mcp_client_factory):
+    name = "Test Suite Probe Skill"
+    description = "A probe skill used by the voice test suite."
+    steps = "Step 1: do the thing. Step 2: verify zebra42 was done."
+
+    saved, err = mcp_client_factory(
+        "skills_mcp.py",
+        "skill_save",
+        {"name": name, "description": description, "steps": steps},
+    )
+    assert not err
+    assert "Saved" in saved
+
+    try:
+        recalled, err = mcp_client_factory("skills_mcp.py", "skill_recall", {"query": "zebra42"})
+        assert not err
+        assert "zebra42" in recalled
+        assert "Steps" in recalled
+    finally:
+        from conftest import docker_exec
+
+        docker_exec("rm", "-f", "/skills/test-suite-probe-skill.md")
+
+
+def test_skill_list(mcp_client_factory):
+    from conftest import docker_exec
+
+    name = "Test Suite List Skill"
+    try:
+        saved, err = mcp_client_factory(
+            "skills_mcp.py",
+            "skill_save",
+            {
+                "name": name,
+                "description": "listing probe skill",
+                "steps": "one step",
+            },
+        )
+        assert not err
+        listed, err = mcp_client_factory("skills_mcp.py", "skill_list", {})
+        assert not err
+        # list output uses the slugified filename
+        assert "test-suite-list-skill" in listed
+    finally:
+        docker_exec("rm", "-f", "/skills/test-suite-list-skill.md")

tests/test_web_ui.py

@@ -0,0 +1,74 @@
+"""Tests for the web UI served by nginx (HTTPS :8090) and the token endpoint."""
+
+import json
+
+import httpx
+
+
+def test_index_serves(client):
+    resp = client.get("/")
+    assert resp.status_code == 200
+    body = resp.text
+    assert "Hope" in body
+    assert "livekit-client.umd.js" in body
+    assert "app.js" in body
+
+
+def test_app_js_serves(client):
+    resp = client.get("/app.js")
+    assert resp.status_code == 200
+    assert "LiveKitClient" in resp.text or "Room" in resp.text
+
+
+def test_style_css_serves(client):
+    resp = client.get("/style.css")
+    assert resp.status_code == 200
+    body = resp.text
+    assert "{" in body and "}" in body
+    assert "color" in body or "background" in body
+
+
+def test_manifest_serves(client):
+    resp = client.get("/manifest.json")
+    assert resp.status_code == 200
+    manifest = json.loads(resp.text)
+    assert "name" in manifest
+
+
+def test_favicon_serves(client):
+    resp = client.get("/favicon.svg")
+    assert resp.status_code == 200
+    assert "svg" in resp.headers.get("content-type", "")
+    assert "<svg" in resp.text
+
+
+def test_livekit_client_bundle_serves(client):
+    resp = client.get("/livekit-client.umd.js")
+    assert resp.status_code == 200
+    assert len(resp.text) > 100_000  # vendored UMD bundle is large
+
+
+def test_token_endpoint_direct():
+    """Token server on 127.0.0.1:8091 (host network mode)."""
+    resp = httpx.post("http://127.0.0.1:8091/token", json={}, timeout=10)
+    assert resp.status_code == 200
+    data = resp.json()
+    assert "token" in data and data["token"].count(".") == 2  # JWT shape
+
+
+def test_token_endpoint_via_nginx(client):
+    """nginx proxies /token to the token server."""
+    resp = client.post("/token", json={})
+    assert resp.status_code == 200
+    data = resp.json()
+    assert "token" in data and data["token"].count(".") == 2
+
+
+def test_livekit_ws_proxy(client):
+    """The /livekit/ path is proxied to the LiveKit server (HTTP :7880)."""
+    resp = client.get("/livekit/")
+    # LiveKit's HTTP API answers on /; a 200 or an API error JSON both prove
+    # the proxy reaches the LiveKit server rather than nginx serving statics.
+    assert resp.status_code in (200, 404, 405)
+    if resp.headers.get("content-type", "").startswith("application/json"):
+        json.loads(resp.text)

web/app.js

@@ -36,6 +36,12 @@ let agentAudioLevel = 0;         // 0..1, smoothed
 let userSpeaking = false;        // local VAD-ish flag from mic level
 let wakeLock = null;
 
+function escapeHtml(s) {
+    const d = document.createElement("div");
+    d.textContent = s || "";
+    return d.innerHTML;
+}
+
 // ── DOM ─────────────────────────────────────────────────────────────────────
 const startBtn = document.getElementById("startBtn");
 const stopBtn = document.getElementById("stopBtn");
@@ -54,6 +60,72 @@ const sheetBackdrop = document.getElementById("sheetBackdrop");
 const closeSheetBtn = document.getElementById("closeSheetBtn");
 const voiceListEl = document.getElementById("voiceList");
 const clearBtn = document.getElementById("clearBtn");
+const tasksBtn = document.getElementById("tasksBtn");
+const tasksPanel = document.getElementById("tasksPanel");
+const tasksList = document.getElementById("tasksList");
+const tasksBadge = document.getElementById("tasksBadge");
+const closeTasksBtn = document.getElementById("closeTasksBtn");
+
+// ── Background tasks state ──────────────────────────────────────────────────
+const tasks = new Map(); // task_id -> {id, description, status, steps: [], result, error}
+
+function updateTasksBadge() {
+    const active = [...tasks.values()].filter(t => t.status === "running" || t.status === "pending").length;
+    if (active > 0) {
+        tasksBadge.hidden = false;
+        tasksBadge.textContent = String(active);
+    } else {
+        tasksBadge.hidden = true;
+    }
+}
+
+function renderTasks() {
+    const items = [...tasks.values()].sort((a, b) => (b.created_at || 0) - (a.created_at || 0));
+    if (items.length === 0) {
+        tasksList.innerHTML = '<p class="tasks-empty">No tasks yet. Ask Hope to research something.</p>';
+        return;
+    }
+    tasksList.innerHTML = "";
+    for (const t of items) {
+        const card = document.createElement("div");
+        card.className = "task-card";
+        const statusClass = t.status || "pending";
+        card.innerHTML = `
+            <div class="task-card-header">
+                <span class="task-status-dot ${statusClass}"></span>
+                <span class="task-desc">${escapeHtml(t.description)}</span>
+            </div>
+            <ul class="task-steps">${t.steps.map(s => `<li class="task-step ${s.role}">${escapeHtml(s.content)}</li>`).join("")}</ul>
+            ${t.result ? `<div class="task-result">${escapeHtml(t.result)}</div>` : ""}
+            ${t.error ? `<div class="task-error">${escapeHtml(t.error)}</div>` : ""}
+        `;
+        tasksList.appendChild(card);
+    }
+}
+
+function handleTaskEvent(data) {
+    const { task_id, event, role, content, result, error, task } = data;
+    if (event === "created" && task) {
+        tasks.set(task_id, { ...task, steps: [] });
+    } else if (event === "step") {
+        const t = tasks.get(task_id);
+        if (t) {
+            t.steps.push({ role, content });
+            t.status = "running";
+        }
+    } else if (event === "completed") {
+        const t = tasks.get(task_id);
+        if (t) { t.status = "completed"; t.result = result; }
+    } else if (event === "failed") {
+        const t = tasks.get(task_id);
+        if (t) { t.status = "failed"; t.error = error; }
+    }
+    updateTasksBadge();
+    renderTasks();
+}
+
+tasksBtn.addEventListener("click", () => tasksPanel.classList.toggle("open"));
+closeTasksBtn.addEventListener("click", () => tasksPanel.classList.remove("open"));
 
 // ── Audio analysers (mic + remote agent audio) ─────────────────────────────
 let micCtx = null;
@@ -352,6 +424,8 @@ function handleDataPacket(payload, participant, topic) {
         const msg = JSON.parse(new TextDecoder().decode(payload));
         if (msg.type === "transcript") {
             addMessage(msg.role || "agent", msg.text);
+        } else if (msg.type === "task_event") {
+            handleTaskEvent(msg);
         }
         // set_voice messages flow the other way; nothing to do client-side.
     } catch (e) {

web/index.html

@@ -64,6 +64,23 @@
         <div class="voice-list" id="voiceList"></div>
     </div>
 
+    <!-- Task panel (background tasks) -->
+    <button id="tasksBtn" class="btn btn-icon tasks-fab" aria-label="Background tasks" title="Background tasks">
+        <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><rect x="3" y="3" width="18" height="18" rx="2"></rect><path d="M9 9h6M9 13h6M9 17h4"></path></svg>
+        <span class="tasks-badge" id="tasksBadge" hidden>0</span>
+    </button>
+    <div class="tasks-panel" id="tasksPanel">
+        <div class="tasks-header">
+            <h2>Background Tasks</h2>
+            <button id="closeTasksBtn" class="btn btn-ghost" aria-label="Close">
+                <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"><path d="M18 6 6 18M6 6l12 12"></path></svg>
+            </button>
+        </div>
+        <div class="tasks-list" id="tasksList">
+            <p class="tasks-empty">No tasks yet. Ask Hope to research something.</p>
+        </div>
+    </div>
+
     <script src="livekit-client.umd.js"></script>
     <script src="app.js"></script>
 </body>

web/style.css

@@ -510,6 +510,178 @@ h1 {
 .transcript::-webkit-scrollbar-thumb,
 .voice-list::-webkit-scrollbar-thumb { background: var(--border); border-radius: 3px; }
 
+/* ── Task panel ──────────────────────────────────────────────────────────── */
+.tasks-fab {
+    position: fixed;
+    bottom: calc(1.5rem + env(safe-area-inset-bottom));
+    right: 1.5rem;
+    z-index: 100;
+}
+
+.tasks-badge {
+    position: absolute;
+    top: -4px;
+    right: -4px;
+    background: #3b82f6;
+    color: #fff;
+    font-size: 0.65rem;
+    font-weight: 700;
+    min-width: 18px;
+    height: 18px;
+    border-radius: 9px;
+    display: grid;
+    place-items: center;
+    padding: 0 4px;
+}
+
+.tasks-panel {
+    position: fixed;
+    top: 0;
+    right: -320px;
+    width: 320px;
+    max-width: 85vw;
+    height: 100dvh;
+    background: var(--panel);
+    border-left: 1px solid var(--border);
+    z-index: 200;
+    display: flex;
+    flex-direction: column;
+    transition: right 0.25s ease;
+}
+
+.tasks-panel.open {
+    right: 0;
+}
+
+.tasks-header {
+    display: flex;
+    align-items: center;
+    justify-content: space-between;
+    padding: 1rem;
+    border-bottom: 1px solid var(--border);
+}
+
+.tasks-header h2 {
+    font-size: 1rem;
+    margin: 0;
+}
+
+.tasks-list {
+    flex: 1;
+    overflow-y: auto;
+    padding: 0.75rem;
+    display: flex;
+    flex-direction: column;
+    gap: 0.75rem;
+}
+
+.tasks-empty {
+    color: var(--text-faint);
+    font-size: 0.85rem;
+    text-align: center;
+    margin-top: 2rem;
+}
+
+.task-card {
+    background: var(--panel-strong);
+    border: 1px solid var(--border);
+    border-radius: 10px;
+    padding: 0.75rem;
+    font-size: 0.82rem;
+}
+
+.task-card-header {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    margin-bottom: 0.4rem;
+}
+
+.task-status-dot {
+    width: 8px;
+    height: 8px;
+    border-radius: 50%;
+    background: #6b7280;
+    flex-shrink: 0;
+}
+
+.task-status-dot.running {
+    background: #3b82f6;
+    animation: pulse-dot 1.5s infinite;
+}
+
+.task-status-dot.completed {
+    background: #22c55e;
+}
+
+.task-status-dot.failed {
+    background: #ef4444;
+}
+
+@keyframes pulse-dot {
+    0%, 100% { opacity: 1; }
+    50% { opacity: 0.4; }
+}
+
+.task-desc {
+    font-weight: 600;
+    color: var(--text);
+    flex: 1;
+    overflow: hidden;
+    text-overflow: ellipsis;
+    white-space: nowrap;
+}
+
+.task-steps {
+    margin-top: 0.5rem;
+    padding-left: 0;
+    list-style: none;
+    display: flex;
+    flex-direction: column;
+    gap: 0.3rem;
+    max-height: 200px;
+    overflow-y: auto;
+}
+
+.task-step {
+    font-size: 0.75rem;
+    color: var(--text-dim);
+    padding: 0.25rem 0.4rem;
+    border-radius: 4px;
+    background: rgba(0,0,0,0.15);
+}
+
+.task-step.tool_call {
+    color: #60a5fa;
+}
+
+.task-step.tool_result {
+    color: var(--text-faint);
+}
+
+.task-step.text {
+    color: var(--text);
+    font-weight: 500;
+}
+
+.task-result {
+    margin-top: 0.5rem;
+    padding: 0.5rem;
+    background: rgba(34, 197, 94, 0.1);
+    border-radius: 6px;
+    font-size: 0.8rem;
+    color: var(--text);
+}
+
+.task-error {
+    margin-top: 0.5rem;
+    padding: 0.5rem;
+    background: rgba(239, 68, 68, 0.1);
+    border-radius: 6px;
+    font-size: 0.8rem;
+    color: #fca5a5;
+}
+
 /* ── Reduced motion ──────────────────────────────────────────────────────── */
 @media (prefers-reduced-motion: reduce) {
     .orb-core,
@@ -517,6 +689,7 @@ h1 {
     .message-row,
     .settings-sheet,
     .sheet-backdrop,
+    .tasks-panel,
     .audio-viz i,
     .mic-meter-fill {
         animation: none !important;

web/token_server.py

@@ -47,7 +47,7 @@ def make_token(room_name: str, identity: str) -> str:
             "canPublishData": True,
         },
         "roomConfig": {
-            "agents": [{"agentName": AGENT_NAME}],
+            "agents": [{"agentName": AGENT_NAME, "jobType": "JT_ROOM"}],
         },
     }
     h = b64url(json.dumps(header).encode())
@@ -66,7 +66,8 @@ class Handler(BaseHTTPRequestHandler):
 
         length = int(self.headers.get("Content-Length", 0))
         body = json.loads(self.rfile.read(length) or b"{}")
-        room_name = "voice-room"
+        # Unique room per session so LiveKit dispatches the agent at creation.
+        room_name = f"voice-{uuid.uuid4().hex[:8]}"
         identity = body.get("identity") or f"user-{uuid.uuid4().hex[:8]}"
 
         token = make_token(room_name, identity)