luke/ai-podcast
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Remove news/research from caller prompts — was causing timeouts

Browse Source 
Research results were bloating the system prompt, making LLM calls
slower and hitting the 20s timeout. Callers don't need news awareness
to have good conversations.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
tcpsyn
2026-02-06 21:44:22 -07:00
parent e45ba2617a
commit 164f8fbc6e
1 changed files with 7 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
backend/main.py
View File

@@ -921,8 +921,8 @@ async def start_call(caller_key: str):
session.start_call(caller_key)
caller = session.caller # This generates the background if needed
if not session.news_headlines:
asyncio.create_task(_fetch_session_headlines())
# if not session.news_headlines:
# asyncio.create_task(_fetch_session_headlines())
return {
"status": "connected",
@@ -1077,7 +1077,7 @@ async def chat(request: ChatRequest):
epoch = _session_epoch
session.add_message("user", request.text)
session._research_task = asyncio.create_task(_background_research(request.text))
# session._research_task = asyncio.create_task(_background_research(request.text))
try:
async with asyncio.timeout(20):
@@ -1091,9 +1091,7 @@ async def chat(request: ChatRequest):
# Include conversation summary and show history for context
conversation_summary = session.get_conversation_summary()
show_history = session.get_show_history()
news_ctx, research_ctx = _build_news_context()
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history,
news_ctx, research_ctx)
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history)
messages = _normalize_messages_for_llm(session.conversation[-10:])
response = await llm_service.generate(
@@ -1622,9 +1620,7 @@ async def _trigger_ai_auto_respond(accumulated_text: str):
conversation_summary = session.get_conversation_summary()
show_history = session.get_show_history()
news_ctx, research_ctx = _build_news_context()
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history,
news_ctx, research_ctx)
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history)
messages = _normalize_messages_for_llm(session.conversation[-10:])
response = await llm_service.generate(
@@ -1674,7 +1670,7 @@ async def _trigger_ai_auto_respond(accumulated_text: str):
broadcast_event("ai_done")
session._research_task = asyncio.create_task(_background_research(accumulated_text))
# session._research_task = asyncio.create_task(_background_research(accumulated_text))
# Also stream to active real caller so they hear the AI
if session.active_real_caller:
@@ -1702,9 +1698,7 @@ async def ai_respond():
conversation_summary = session.get_conversation_summary()
show_history = session.get_show_history()
news_ctx, research_ctx = _build_news_context()
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history,
news_ctx, research_ctx)
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history)
messages = _normalize_messages_for_llm(session.conversation[-10:])
response = await llm_service.generate(