<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Hacker News: divyanshgolyan</title><link>https://news.ycombinator.com/user?id=divyanshgolyan</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Wed, 18 Nov 2026 15:35:49 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=divyanshgolyan" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[Show HN: Rui, orchestrating 100 agents in 2.5 MB of RAM]]></title><description><![CDATA[
<p>Hi HN, I've built Rui (cotton in hindi), an agent runtime built for memory efficiency.<p>The main design choice is that sessions share a host rather than each being a process. Agents are state in DB and the host handles the active work through a share event loop.<p>LLMs are getting better at coordination and teams (swarms) of agents are coming. Memory is getting expensive and my machine is always using swap. We cannot be wasting hundreds of MBs of RAM for an orchestrator (AKA the harness) of LLMs and tools.<p>More information on the architecture and measurement methodology is at <a href="https://divyanshgolyan.com/posts/hundred-agents/" rel="nofollow">https://divyanshgolyan.com/posts/hundred-agents/</a><p>I'd love feedback about this design boxing me into a corner and any large tradeoffs I'm not seeing!</p>
<hr>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=50403797">https://news.ycombinator.com/item?id=50403797</a></p>
<p>Points: 1</p>
<p># Comments: 0</p>
]]></description><pubDate>Sat, 07 Nov 2026 12:02:28 +0000</pubDate><link>https://github.com/DivyanshGolyan/rui</link><dc:creator>divyanshgolyan</dc:creator><comments>https://news.ycombinator.com/item?id=50403797</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50403797</guid></item></channel></rss>