How to trigger a new search and/or flush all search results within a chat

[Scott8586]

I'm finding that the model will cache a search result, and so fails when I ask about a different device.

1. Start a new chat
2. I ask it about lights, I can see it uses ha-mcp to go and get entities, and return values. All good.
3. When I then ask for disk drive temperatures in the same conversation, I can see it's not searching for this new information, it apparently looks at the cached search results, and claims it can't find anything.

If I start a new chat to look for disk drive temperatures then it starts a new search, I can see the results, and it works.

Within a chat, I've tried saying "with a new search", or "flush the search results", but no joy - it seems to require a new chat to start a new search.

How can I trigger the model to flush the old search and start a new search within a chat?

Interface: Open WebUI
Model: ministral-3:7b
ha-mcp environment:

ENABLED_TOOL_MODULES="smart_search,tools_search,tools_service,tools_services,tools_system"

[Scott8586]

Ok, it looks like I can answer my own request, clear or flush doesn't seem to work but something along the lines of:

search for any temperature sensors using “_temperature” and summarize

Instead of:

show me....

appears to start a new search. Though it sure would be nice to have "code" to tell the model to force a new search.

[sergeykad]

This is a known limitation of smaller models. They see prior search results in context and reuse them instead of calling the tool again — even when the new question needs different entities. Larger models understand that "disk drive temperatures" requires a different search than "lights"; smaller models don't.

Your best workaround: be explicit. Say "search for X using ha_search_entities" rather than "show me X." You can also add to your system prompt: "Always perform a new entity search for each request; do not reuse previous results."

There's no way to force tool calls server-side.