Show more specific warning to llm on duplicate tool use during research

This commit is contained in:
Debanjum
2025-08-15 14:17:35 -07:00
parent caf0b994e8
commit 4274f58dbd
2 changed files with 2 additions and 2 deletions

View File

@@ -197,7 +197,7 @@ async def apick_next_tool(
if i.warning is None and isinstance(i.query, ToolCall)
}
if (parsed_response.name, dict_to_tuple(parsed_response.args)) in previous_tool_query_combinations:
warning = "Repeated tool, query combination detected. Skipping iteration. Try something different."
warning = f"Repeated tool, query combination detected. You've already called {parsed_response.name} with args: {parsed_response.args}. Try something different."
# Only send client status updates if we'll execute this iteration and model has thoughts to share.
elif send_status_func and not is_none_or_empty(response.thought):
async for event in send_status_func(response.thought):

View File

@@ -588,7 +588,7 @@ tools_for_research_llm = {
description=dedent(
"""
To have the tool AI semantic search through the user's knowledge base.
Helpful to answer questions for which finding some relevant notes or documents can complete the search. Example: "When was Tom born?"
Helpful to answer questions for which finding some relevant notes or documents can be useful. Example: "When was Tom born?"
This tool AI cannot find all relevant notes or documents, only a subset of them.
It is a good starting point to find keywords, discover similar topics or related concepts and some relevant notes or documents.
For a given query, the tool AI can perform a maximum of {max_search_queries} semantic search subqueries per iteration.