<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Attestor writing</title>
    <link>https://attestor.dev/blog/</link>
    <atom:link href="https://attestor.dev/blog/feed.xml" rel="self" type="application/rss+xml"/>
    <description>Essays on governed memory for multi-agent AI systems: who is allowed to write, who is allowed to forget, and how to prove it afterwards.</description>
    <language>en</language>
    <lastBuildDate>Fri, 18 Sep 2026 00:00:00 GMT</lastBuildDate>
    <item>
      <title>Attestor is no longer memory for your coding assistant</title>
      <link>https://attestor.dev/blog/no-longer-memory-for-your-coding-assistant.html</link>
      <guid isPermaLink="true">https://attestor.dev/blog/no-longer-memory-for-your-coding-assistant.html</guid>
      <pubDate>Fri, 18 Sep 2026 00:00:00 GMT</pubDate>
      <description>The editor vendors shipped memory for free. The benchmark race became a race about which LLM answers the question. And every serious conversation was about something else entirely: when many agents share one memory, who is allowed to write, who is allowed to forget, and can you prove it afterwards. So that is what Attestor is now: governed memory for multi-agent production meshes. What that means in the code, not the deck.</description>
    </item>
  </channel>
</rss>
