<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Cogni updates</title>
  <subtitle>Releases and measured improvements to Cogni.</subtitle>
  <link href="https://getcogni.io/updates.xml" rel="self" />
  <link href="https://getcogni.io/updates" />
  <id>https://getcogni.io/updates</id>
  <updated>2026-08-12T00:00:00Z</updated>
  <author><name>Cogni</name><email>support@getcogni.io</email></author>

  <entry>
    <title>Chain recall got substantially better on the default settings</title>
    <link href="https://getcogni.io/updates#2026-08-12-recall" />
    <id>https://getcogni.io/updates#2026-08-12-recall</id>
    <updated>2026-08-12T00:00:00Z</updated>
    <category term="Recall" />
    <summary>Graph traversal seeds are now bounded to half the response, so the traversal's results reach the caller. Measured through the real server on the default mode: 0.06 to 0.48 on cross-vocabulary multi-hop questions, at 26% fewer internal searches. Retrieval breadth now also scales with store size.</summary>
  </entry>

  <entry>
    <title>Team spaces are self-serve</title>
    <link href="https://getcogni.io/updates#2026-08-12-team-spaces" />
    <id>https://getcogni.io/updates#2026-08-12-team-spaces</id>
    <updated>2026-08-12T00:00:00Z</updated>
    <category term="Teams" />
    <summary>Creating a team space, adding and removing members, and viewing the roster now all happen from your account page. Administration is limited to a space's owners, and the last owner cannot be removed.</summary>
  </entry>

  <entry>
    <title>Cogni is live</title>
    <link href="https://getcogni.io/updates#2026-08-10-live" />
    <id>https://getcogni.io/updates#2026-08-10-live</id>
    <updated>2026-08-10T00:00:00Z</updated>
    <category term="Launch" />
    <summary>Sign-up is open. Cogni is an MCP memory server that gives any model a memory which persists between sessions and follows the connections between what you have told it. One memory across every client you sign in from. No vector database, no GPU, and no LLM in the retrieval path.</summary>
  </entry>

</feed>
