<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Jonah T</title>
    <description>The latest articles on DEV Community by Jonah T (@dogtorjonah).</description>
    <link>https://dev.to/dogtorjonah</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3975118%2Fe2871b3d-ee68-4646-a0d5-52d168ada7de.png</url>
      <title>DEV Community: Jonah T</title>
      <link>https://dev.to/dogtorjonah</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/dogtorjonah"/>
    <language>en</language>
    <item>
      <title>Context Warp Drive: deterministic folding for long-running LLM agents</title>
      <dc:creator>Jonah T</dc:creator>
      <pubDate>Mon, 29 Jun 2026 20:50:05 +0000</pubDate>
      <link>https://dev.to/dogtorjonah/context-warp-drive-deterministic-folding-for-long-running-llm-agents-2jok</link>
      <guid>https://dev.to/dogtorjonah/context-warp-drive-deterministic-folding-for-long-running-llm-agents-2jok</guid>
      <description>&lt;p&gt;Context Warp Drive is an open-source TypeScript library for keeping long-running LLM agents under the context ceiling without asking another model to summarize their state.&lt;/p&gt;

&lt;p&gt;The core trick is deterministic folding. Instead of summarization calls, it compacts old transcript regions into stable fold artifacts, preserves exact coordinates for recall, and keeps recent/live messages append-only. That gives a harness a smaller prompt while still leaving a trail back to the original material.&lt;/p&gt;

&lt;p&gt;Why I built it:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Agents die weirdly when context pressure gets high.&lt;/li&gt;
&lt;li&gt;Provider prompt caches reward byte-stable prefixes.&lt;/li&gt;
&lt;li&gt;Summaries are useful, but they are not the same thing as deterministic state.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;What is in the repo today:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Rolling fold compaction with Coordinate Closet references&lt;/li&gt;
&lt;li&gt;Cache-hot freeze for provider prefix caching&lt;/li&gt;
&lt;li&gt;Ambient recall/page-in for relevant older material&lt;/li&gt;
&lt;li&gt;Durable episodic memory extraction&lt;/li&gt;
&lt;li&gt;Register glyphs for continuity signals&lt;/li&gt;
&lt;li&gt;Provider helpers for Anthropic caching, with OpenAI/Gemini-shaped adapters in the package&lt;/li&gt;
&lt;li&gt;459 deterministic tests in the current tree&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Current status:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;GitHub: &lt;a href="https://github.com/dogtorjonah/context-warp-drive" rel="noopener noreferrer"&gt;https://github.com/dogtorjonah/context-warp-drive&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Hacker News discussion: &lt;a href="https://news.ycombinator.com/item?id=48722788" rel="noopener noreferrer"&gt;https://news.ycombinator.com/item?id=48722788&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;npm package metadata/build/tests are ready for &lt;code&gt;context-warp-drive@0.1.0&lt;/code&gt;, but the actual npm publish is waiting on account auth.&lt;/li&gt;
&lt;li&gt;JSR needs a jsr.json/Deno config before it is a clean target.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The thing I want feedback on is the shape of the primitive. Is deterministic folding the right lower-level abstraction for long-lived agents, or should this sit higher up as app-specific memory? I am especially interested in people trying to break the cache assumptions, fold invariants, or recall behavior.&lt;/p&gt;

</description>
      <category>agents</category>
      <category>llm</category>
      <category>opensource</category>
      <category>typescript</category>
    </item>
    <item>
      <title>I built a browser Mandelbrot explorer for deep zooms</title>
      <dc:creator>Jonah T</dc:creator>
      <pubDate>Tue, 09 Jun 2026 03:43:08 +0000</pubDate>
      <link>https://dev.to/dogtorjonah/i-built-a-browser-mandelbrot-explorer-for-deep-zooms-5e86</link>
      <guid>https://dev.to/dogtorjonah/i-built-a-browser-mandelbrot-explorer-for-deep-zooms-5e86</guid>
      <description>&lt;p&gt;After a 7-week build sprint, I launched Zooming Fractal: &lt;a href="https://zoomingfractal.com" rel="noopener noreferrer"&gt;https://zoomingfractal.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;It's a free in-browser Mandelbrot set explorer. You can zoom deep into the fractal, tune colors and detail, and save or share exact locations as URLs.&lt;/p&gt;

&lt;p&gt;No signup, no install, no paywall. I mostly wanted it to feel like a little visual instrument: open it, wander, find a weird beautiful region, send that exact view to someone.&lt;/p&gt;

&lt;p&gt;I'd love feedback from DEV folks on the first-use experience:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Is it obvious what to do in the first 10 seconds?&lt;/li&gt;
&lt;li&gt;Does it feel smooth on your browser/device?&lt;/li&gt;
&lt;li&gt;Would anything make the share-link flow more useful?&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Try it here: &lt;a href="https://zoomingfractal.com" rel="noopener noreferrer"&gt;https://zoomingfractal.com&lt;/a&gt;&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>webdev</category>
      <category>javascript</category>
      <category>creativecoding</category>
    </item>
  </channel>
</rss>
