Most AI chats forget everything when you close the tab. Every new conversation starts from zero — the AI doesn't know who you are, what you're working on, what you decided last week, or what you've taught it. You type the same context again and again.
Wick is different. Wick keeps a small memory/ folder on your computer — plain markdown files that store what she's learned about you. Who you are. What you're building. Decisions you've made. Predictions you've logged. Corrections you've given her. Over time, the files grow. Sessions compound. Session 30 with Wick is meaningfully better than session 1, because she actually remembers you.
Free. Open source. Nothing phones home. Built for Claude Code. Other tools that read AGENTS.md, or take a pasted system prompt, can run her too, and the README says what each one actually reads. The memory/ folder is plain files: it moves with you between tools and machines.
And she doesn't just remember. She says what she doesn't know: that rule comes first, and it covers the documents you hand her as well as what she knows. She keeps a prediction ledger that a program scores, beside no-skill baselines, so you can see whether your confidence is earned. And she notices repetition: a task you keep repeating can become a script or a skill.
We test the rules before they ship. In November 2026 we measured, for the first time, the file Wick loads every session. On Claude Opus 5.5 the old file made the model worse at admitting when a passage did not hold the answer. Version 2.0 is a quarter of the size and removes that harm. It does not beat one plain sentence telling a model to say "I don't know", and we say so. The test and the numbers are in the repository.