<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
<channel>
  <title>Frank Zhang — Notes</title>
  <link>https://frankzhang.ca/notes/</link>
  <atom:link href="https://frankzhang.ca/notes/feed.xml" rel="self" type="application/rss+xml"/>
  <description>Engineering notes by Frank Zhang.</description>
  <language>en</language>
  <lastBuildDate>Tue, 27 Oct 2026 00:00:00 GMT</lastBuildDate>
  <item>
    <title>Agent Memory Is Easy to Store. Trustworthy Recall Is Harder.</title>
    <link>https://frankzhang.ca/notes/agent-memory-trustworthy-recall/</link>
    <guid isPermaLink="true">https://frankzhang.ca/notes/agent-memory-trustworthy-recall/</guid>
    <pubDate>Tue, 27 Oct 2026 00:00:00 GMT</pubDate>
    <category>Agent Systems</category>
    <description>The architecture and evidence behind explicit memory, trustworthy recall, and a decision to defer automatic capture in Agent Memory Bridge.</description>
    <content:encoded><![CDATA[<p>An agent can save a sentence in a database. The harder question is whether a later session should use it—and what should happen before that sentence is saved at all.</p>
<p>Agent Memory Bridge (AMB) is my open-source project for carrying project knowledge across agent sessions. It grew from a practical problem: decisions and their reasons get lost in conversations, but blindly replaying old context can bring back stale assumptions, unverified fixes, or sensitive material.</p>
<h2 id="remember-the-reason-not-the-whole-conversation">Remember the reason, not the whole conversation</h2>
<p>AMB's local-first model gives an agent a way to record an explicit decision and its context, then recall relevant knowledge in a later session. Its Model Context Protocol (MCP) interface exposes memory operations to compatible clients. The store keeps revision relationships so a corrected decision can supersede an older one rather than quietly erasing its history.</p>
<p>That does not make a stored claim true. Provenance helps a reader find where a claim came from; it is not proof of correctness or authenticated identity. A later agent still needs to compare recalled guidance with current code and evidence. Memory explains why a choice was made; the live repository shows what exists now.</p>
<blockquote>
<p>Recall is a starting point for a check, not a substitute for one.</p>
</blockquote>
<p>For example, a team might record why a project uses one authoritative store. A later session can retrieve that reason before changing storage—but must still inspect the current implementation and ask whether the original constraints remain valid.</p>
<h2 id="convenience-creates-a-new-boundary">Convenience creates a new boundary</h2>
<p>Explicit memory has a cost: someone or something must decide what is worth retaining. In <a href="https://github.com/zzhang82/Agent-Memory-Bridge/issues/72">research issue #72</a>, I investigated whether lower-friction capture could improve that experience without silently promoting unverified material into trusted guidance.</p>
<p>There are several different steps: observing a source, extracting a candidate, storing it, reviewing it, and promoting it to trusted guidance. They are not interchangeable. AMB supports bounded, structured capture into a review lane; that is not the same as automatically extracting useful knowledge from ordinary conversations. A candidate can be stored even when it is not returned by ordinary recall.</p>
<p>The distinction matters because review controls authority, not confidentiality. If sensitive material is persisted before review, keeping it out of recall does not undo the write.</p>
<h2 id="the-check-that-failed">The check that failed</h2>
<p>The <a href="https://github.com/zzhang82/Agent-Memory-Bridge/issues/72#issuecomment-6380224722">published #72 report</a> describes an isolated synthetic privacy check. A known key-pattern input was rejected before storage, but an opaque synthetic identifier designated sensitive by the fixture reached a hidden review candidate. The no-storage check <strong>failed</strong>. No real secret or production database was used; this was not evidence of a production leak.</p>
<p>It was also not a solved security issue. A pattern match can catch known shapes without recognizing every sensitive value. A caller-provided safety label is not proof of confidentiality. Other passing tests in the report did not erase this negative result or establish protection against arbitrary sensitive content.</p>
<p>The published material includes a <a href="https://github.com/zzhang82/Agent-Memory-Bridge/issues/72#issuecomment-6381047371">redacted evidence capsule</a>; it does not expose the complete raw run package. The result is historical evidence tied to the research snapshot, not a claim about every version or deployment.</p>
<h2 id="what-i-chose-not-to-build">What I chose not to build</h2>
<p>I deferred broader automatic capture rather than treating a review queue as a privacy guarantee. The follow-up <a href="https://github.com/zzhang82/Agent-Memory-Bridge/issues/73">#73 design research</a> examines approved sources, minimization before writing, retention surfaces, and the limits of detecting opaque sensitive data. Its <a href="https://github.com/zzhang82/Agent-Memory-Bridge/issues/73#issuecomment-6382268385">closeout</a> does not authorize an extractor or automatic persistence; proposed synthetic experiments remained not run.</p>
<p>That decision trades convenience for a more explicit boundary. Users must still identify valuable knowledge and request recall. It also leaves difficult questions open: which sources are safe to inspect, which host or provider may retain a copy, and whether the benefit justifies the operational and review cost. A future experiment would need separate authorization and evidence across those boundaries.</p>
<h2 id="what-the-evidence-does-and-does-not-show">What the evidence does—and does not—show</h2>
<p>The project provides a concrete path for storing and recalling explicit project knowledge, with revisions and governance. The research also shows why a successful candidate or a passing known-pattern test is not enough to claim safe automatic learning.</p>
<p>A <a href="https://github.com/zzhang82/Agent-Memory-Bridge/issues/72#issuecomment-6380809012">small synthetic comparison</a> found that a standing recall procedure mattered in the tested sessions. Its sample and host scope were limited; it did not prove broad productivity gains or that packaging alone caused the result. The engineering lesson is narrower: make the boundary visible, test a plausible failure, and preserve the failure when it changes the design.</p>
<h2 id="sources-and-further-reading">Sources and further reading</h2>
<ul>
<li><a href="https://github.com/zzhang82/Agent-Memory-Bridge/blob/main/README.md">AMB documentation and current capabilities</a></li>
<li><a href="https://github.com/zzhang82/Agent-Memory-Bridge/blob/main/docs/WRITE-SIDE-CAPTURE.md">Write-side capture and review boundaries</a></li>
<li><a href="https://github.com/zzhang82/Agent-Memory-Bridge/issues/72#issuecomment-6380994237">Issue #72 research closeout</a> and <a href="https://github.com/zzhang82/Agent-Memory-Bridge/issues/72#issuecomment-6380224722">historical test report</a></li>
<li><a href="https://github.com/zzhang82/Agent-Memory-Bridge/issues/73">Issue #73: pre-write minimization research</a></li>
</ul>]]></content:encoded>
  </item>
</channel>
</rss>
