2026-06-03 17:26 UTC
@b0rk@social.jvns.ca I feel you re: mixed feeling. Even when everything works great, I see how many searches ran automatically and I wonder how sustainable that can be in the long term…
Feeding search results into a model is def just another type of RAG, what I like though is that agents autonomously evaluate and delve deeper if needed
(or call other tools like here: https://pi.dev/session/#60456692bce68a9b2611fbfcf60fc808).
Direct embedding search (like #BYOTA) is also cool, but depends on what you can/want to index (fun challenge tho😁)
Replies (1)
-
@mala@fosstodon.org 2026-06-03 17:26
@b0rk@social.jvns.ca one thing I love of SearXNG is how you can configure different search engines per task. If I had to bet on some new cool agentic search, it would be on something that makes use of similar metadata to automatically run searches on sets of custom, ad-hoc engines. And oh yes, there’s knowledge graphs too: wikidata has recently shared an MCP server / API I’d love to play with (https://wd-mcp.wmcloud.org/ )💙