#Are there any (free) LLM conversation agents which will actually search the internet for answers?

1 messages · Page 1 of 1 (latest)

rapid cipher
#

Currently using Google Gemini API integration and it's getting rather annoying whenever you ask a lot of things of it it implies it doesn't know because it doesnt have access to the internet etc. Are there any conversation agents which would be able to go off and search the internet? (I assume ChatGPT can, but I cannot justify paying £20 a month specifically for this usecase, it would be cheaper and easier to just use alexa+ whenever that comes out... which frankly I really don't really want to do).

I've got ollama, but its running off of CPU on my server, so its pretty unusably slow unless I use a tiny LLM model, which I beleive the small ones don't have support for "tools" which is what they need to be able to search the internet and so on.

#

Are there any (free) LLM conversation agents which will actually search the internet for answers?

sharp kindle
#

Check out MCP

#

Basically to use Internet search you need tools exposed to the llm it can use to retrieve the data

rapid cipher
#

ah great ok, thanks. I'll look into getting that set up on my server as a container. hopefully it doesnt add to much additional delays on getting answers from the voice assistant 🙂

rapid cipher
#

hmm can't seem to get it working . installed the apps as instructed (couldnt find a container which did it) and it says its running but the integration can't connect to it 😦

unreal osprey
#

You don’t need 20 bucks a month for ChatGPT by the way. At least I just pay per query and it’s fairly cheap (when comparing to power costs in my country at least).

#

I spent 14 cents for the whole of February to OpenAI for reference. Only have one PE but it’s used fairly regularly.

sharp kindle
#

Yeah i use gpt-4o-mini every day across 4 voice pe devices and it's rare that i spend more then a few bucks a month on the api

#

That said, it doesn't do Internet searches either,b still need tools 🙂

rapid cipher
#

oh really, cool. I assumed that you had to pay for "pro" or whatnot to get the API access. would that mean that i dont need to do this MCP intermediary? (which i'm still struggling to get up and running)

edit: ah sorry just re-read Nicks reply. i'll crack on trying to get it working then 🙂

rapid cipher
#

ok i have it al running with a mcp-proxy now. but it seems that I have to specify each time for it to check a specific website for things, rather than it having the smarts to simply google the answers etc much like chatgpt does nowdays. am i completely misunderstanding the use of this mcp-proxy or is there a bunch of other stuff i need to do for it to work in the manner which is needed.

sharp kindle
#

Not sure if the models know to go to the Internet without being prompted to do so. That would probably use their own internal knowledge graph to answer questions. Maybe the system prompt could have something added to direct it to use the search tool. Also not sure the llm would just search the Internet, probably need a Google search api and something like RAG to achieve something like that

rapid cipher
#

yeah I don't think using this is going to work at the moment as it is. it doesn't go off and do any searching of it's own and also if you have it set up to use this mcp proxy it means you can't use the assistant to do anything else (any normal questions which it shouldn't need to search the internet for... or even control your home). so it makes it pretty pointless as it is