March 25, 2026
AI dreams, devs scream
Show HN: A plain-text cognitive architecture for Claude Code
Plain‑text AI memory drops; devs love the idea, roast the write‑up, and yell “is this new”
TLDR: Cog gives Claude Code a plain‑text “memory” and a nightly cleanup so it remembers you. Commenters love the simple idea but roast the bot-sounding docs, debate whether it’s new versus other memory tools, and worry about unreliable long‑term notes—even as power users share clever workflows.
Meet Cog: a DIY “memory” for Claude Code that lives in plain‑text files, not some mysterious database. You write simple rules in Markdown, the bot files its own notes, and each night it does REM sleep—a cleanup pass that trims, rewrites, and improves itself, with every change visible in the git log. The demo screams transparency and control.
But the comments? Spicy. One camp loves the concept and hates the voice. “Bot wrote the docs” energy had readers begging for a human write‑up, with kixiQu calling the style “so painful” it overshadowed the tool. Another camp rolled in with the déjà vu alarms: CharlesW asked how this differs from Superpowers’ episodic memory or Anthropic’s Auto Dream. Is Cog clever branding for a familiar trick, or actually new?
Meanwhile, rodspeed hit a nerve: memory rot. Old guesses and half‑truths age badly, and nothing degrades gracefully—do we trust yesterday’s hunch like today’s fact? Others traded recipes: Real_Egor pointed to Google’s “Antigravity” temp files for inspiration (brain, conversations, artifacts, annotations), while K0balt shared an onboarding/shutdown ritual with journals and scratchpads. The memes wrote themselves: “AI does REM, we do RTFM,” “git log as therapy notes,” and “the filesystem is the new UI.” Whether Cog is revolution or remix, the crowd agrees on one thing: give the bot a diary, but make it readable—and reliable.
Key Points
- •Cog introduces persistent memory for Claude Code using plain text files instead of a database.
- •Rules are defined in markdown; the model builds structure around those rules and follows them.
- •A nightly pipeline consolidates conversations, extracts patterns, prunes stale data, and rewrites rules.
- •All decisions and rule changes are transparent and editable, with changes tracked in a git log.
- •The project is an experimental learning tool to study how AI organizes knowledge and benefits from self-reflection.