Should Perplexica switch to function calling? #814
Replies: 3 comments 2 replies
-
|
Sounds great to me! Available new models aren't getting worse yet so compatibility doesn't worry me any, they're still getting better. |
Beta Was this translation helpful? Give feedback.
-
|
Tool/function calling (maybe even for "deep research" purposes), and then MCP-ify Perplexica further so they can work well with IDEs and agentic tools + Memory/Graphs. Kinda fever dream concept of tool-to-API |
Beta Was this translation helpful? Give feedback.
-
|
Would be awesome if we could have MCP. I am midst installing an MCP server for RAG - that would became usable with Perplexica too. :) Projects like Perplexica and co are really necessary because of data security and costs. My guess is, that the pro user tier for OpenAI and co won't stay at 10 or 20 bucks.... So, thanks for Perplexica! :) |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Hey Everyone,
Perplexica has been around for a while now, and I think it’s time to completely revamp the old search handling system. Right now, the query goes to the LLM to decide whether searching is needed or not, and then based on that, either a search is done or the answer is generated directly. It’s slow, buggy at times, and honestly not that reliable.
When I first built Perplexica, tool/function calling models were limited (both in quanity and quality). But now, things have changed. Most of the newer models, even the smaller ones, support function/tool calling properly.
So I’m planning to switch to a function-calling based system for search handling. This would make Perplexica way more reliable and significantly boost answer quality. It’ll also allow us to do a lot more cool stuff directly inside the chat, like showing weather info, live maps for location-based queries, previews of local businesses, shopping mode, calendar/event suggestions, and more stuff like that which just isn’t possible with the current setup.
But yeah, the catch is that models which don’t support function calling will stop working with Perplexica. Only models that support it will be usable going forward.
Please cast your votes below.
24 votes ·
Beta Was this translation helpful? Give feedback.
All reactions