AI-powered search & scraping | 16x more affordable API, built for devs and AI agents Explore at https://xitter.cf/t.co/L3W1RHE1Js | SN22 on Bittensor
⛓️
Joined January 2025
- Tweets703
- Following76
- Followers1.7K
- Likes807
A RAG app is only as good as the documents behind it.
Give this script a topic and 3 to 5 sub-questions. It keeps the main article text, drops navigation, footers and cookie banners, and splits it into chunks that follow the page's headings.
Link on LinkedIn.
Before you pitch, read what they already said about your space.
Give this script 10 to 15 investor or fund accounts on X and one topic. You get their public posts on it, newest first, with activity in the last 30 days flagged.
Link on LinkedIn.
A short digest of today's news from the web and X, with a link behind every line.
This script turns numbered citations into real links and tags older or undated sources.
It can post it to Telegram. Link on LinkedIn.
Hundreds of papers land on arXiv every week. Which ones are people actually talking about?
This script checks each paper against the public arXiv API for its real title, authors and date, then finds the X conversation around it.
Link on LinkedIn.
A good API demo shows more than results.
The Desearch Web Search demo makes cost and cache state visible, labels saved examples, and pauses instead of presenting fallback data as live.
Try it:
web-search.demos.desearch.ai
#Desearch #AISearch #AIAgents
Your stars jumped. Which post did it?
This script gives you a day-by-day view of the public X posts and web pages that mention your repo. It searches X by repo URL and by name, with Latest and Top passes.
Link on LinkedIn.
Embedding tasks are part of SN22 main. If you have a GPU, you can now embed the pages the network crawls with Desearch-Embedding-4B.
Validators embed a sample again and compare. Tasks that pass, or are not picked for a check, are paid by the characters embedded.
On Oct 1 we showed hosted MCP. Here is an agent built on it.
It connects, lists tools, calls them, reads the top 3 pages as plain text and answers with sources and a cost log.
The default loop is scripted and needs no LLM. Link on LinkedIn.
Small Python scripts, each one a real job your agent can do.
For developers, marketers and analysts. Each one uses the Desearch API and links to its source.
One a day for the next few weeks. Link on LinkedIn.
One agent task: "What has been published about open source AI agent frameworks since September 1?"
One Desearch web search call with a start date and page text. Every result came back inside the date range, with the full page text.
#searchAPI
Give Desearch crawl a URL it can't fetch and it now tells you plainly that the URL is invalid.
Less guessing when your agent hits a bad link.
#searchAPI
