fix: delay and vary spoken wait notices

agent/hermes_llm.py

@@ -11,6 +11,7 @@ from __future__ import annotations
 import asyncio
 import json
 import math
+import random
 import uuid
 from collections.abc import AsyncIterator, Awaitable, Callable
 from dataclasses import replace
@@ -24,7 +25,13 @@ from livekit.agents.types import (
 
 ProgressCallback = Callable[[dict[str, Any]], Awaitable[None]]
 WaitNoticeCallback = Callable[[str], Awaitable[None]]
-WAIT_NOTICE = "I'm working on that. I'll let you know when it's ready."
+WAIT_NOTICE_SECONDS = 12.0
+WAIT_NOTICES = (
+    "I'm still working on that. I'll let you know when it's ready.",
+    "This is taking a little longer. I'm on it.",
+    "I'm still looking into that. I'll let you know what I find.",
+    "I need a little more time on this. I'm still on it.",
+)
 
 
 class HermesLLMError(APIError):
@@ -119,6 +126,7 @@ class HermesLLM(lk_llm.LLM):
         }
         self._on_progress = on_progress
         self._on_wait_notice = on_wait_notice
+        self._last_wait_notice: str | None = None
         self._on_answer_start = on_answer_start
         self._timeout_seconds = timeout_seconds
         self._owns_client = http_client is None
@@ -176,10 +184,17 @@ class HermesLLMStream(lk_llm.LLMStream):
         if self._noticed or self._has_answer or self._hermes._on_wait_notice is None:
             return
         self._noticed = True  # latch before awaiting, including competing timer
-        await self._callback(self._hermes._on_wait_notice, WAIT_NOTICE)
+        notice = random.choice([
+            text for text in WAIT_NOTICES if text != self._hermes._last_wait_notice
+        ])
+        self._hermes._last_wait_notice = notice
+        await self._callback(self._hermes._on_wait_notice, notice)
 
     async def _latency_notice(self) -> None:
-        await asyncio.sleep(3.0)
+        # Status/tool-start events are not predictions of long latency: Hermes
+        # emits them even for ordinary replies. Only elapsed time without an
+        # answer warrants spoken feedback; UI progress remains independent.
+        await asyncio.sleep(WAIT_NOTICE_SECONDS)
         if not self._has_answer:
             await self._progress("working", "Working on your request.")
             await self._notice()
@@ -244,7 +259,6 @@ class HermesLLMStream(lk_llm.LLMStream):
                     if item.get("status") in ("failed", "error"):
                         raise HermesLLMError("upstream_error")
                     await self._progress("working", "Working on your request.")
-                    await self._notice()
                     continue
                 if event == "hermes.tool.progress":
                     status = item.get("status", "working")
@@ -255,8 +269,6 @@ class HermesLLMStream(lk_llm.LLMStream):
                         name = item.get("tool_name", item.get("tool", item.get("name")))
                         message = f"Working on {name}." if isinstance(name, str) else "Working on your request."
                     await self._progress("working", message)
-                    if status not in ("completed", "complete", "success", "finished"):
-                        await self._notice()
                     continue
                 choices = item.get("choices", [])
                 if not isinstance(choices, list):