All insights
Small BusinessOpen SourceAutomationTry This

Your AI Agent Forgets Everything. This Tiny Script Fixes That.

Jesse Burcsik·July 31, 2026·3 min read

Every time you open a new chat with your AI agent, it starts from scratch. No memory of the client preferences you explained last week, no recollection of the workflow you figured out together, no idea you prefer bullet points over paragraphs. You start over. Again.

What's happening

Victor Taelin, an indie developer known for building small, sharp tools, released OptMem this week: a free, open-source memory system for AI agents built from a few hundred tokens of prompt and a single Python script.

The idea is beautifully simple. You paste a memory block into your agent's config file (AGENTS.md or CLAUDE.md, depending on what you use). From then on, your agent has access to a small set of memory commands. The agent calls memo note to record something worth keeping, memo recall to search its notes, and memo nap to compress older memories. The database is an append-only log. Nothing is ever deleted. The memory stays a constant 208 lines no matter how long you use it, with older details fading into compressed summaries while recent ones stay verbatim.

The repo hit GitHub Trending this week and picked up nearly 600 stars in a few days. It works with any agent that reads a standard config file: Claude Code, OpenClaw, Codex CLI, Hermes Agent, and most others. No subscription, no server, no account required.

Try this this week

  • Head to github.com/VictorTaelin/OptMem and read the README (it is one page)
  • Follow the setup: download the memo script and add it to your PATH (the README has exact steps for Mac, Linux, and Windows)
  • Paste the Memory block from the README into your AGENTS.md or CLAUDE.md
  • Start your next work session. At any point, ask your agent to run memo note with a short detail: a client preference, a recurring pattern, something you had to explain more than once
  • Close the session, reopen, and type memo wake. The agent will read back what it remembers before you start

Give it a week of real work before you judge it. The first session feels like nothing changed. By day five, your agent knows your rhythm without being told.

The bigger picture

The real leverage of AI agents for a small team is not speed on a single task. It is the knowledge that stacks up over weeks of working together. Every preference you do not have to re-explain and every mistake your agent does not repeat twice adds up to something real. Most commercial tools want to sell you that memory as a monthly subscription. OptMem is a text file and a short script. It is rough around the edges, but it tackles a genuine friction point, and the setup takes about fifteen minutes. Start here, see how much it changes a week of work, then decide if you want something fancier. That is always the right order.

Like this? Let's put it to work.

We help small teams turn ideas like these into working systems, no six-month roadmap required.