Grok Build now keeps project memory: notes after each turn, /memory to read them
Background markdown on conventions, decisions and facts. Per-project plus global preferences. /dream files the notes. Secrets, task state and what’s already in the repo stay out.
Feature1 minEasy
Priya Shah1 min read
- xAI
- Grok Build
- Memory
- Agents
- Developers

Photo: xAISource page
Grok Build, xAI’s terminal coding agent, now writes durable notes after each completed turn and reads them back in later sessions. The company posted the change on Wednesday. Capture runs in the background and does not block the session. What it stores, xAI said, is conventions, decisions, and project facts — how the suite is run, where a subsystem lives, why a choice was made. Notes are markdown, one topic per subject. Each project has its own workspace; a global scope holds preferences that apply everywhere. Instructions in the current conversation still win over a note.
Two commands sit on top of the files. `/memory` opens a read-only browser of every memory file, grouped by scope. `/dream` folds recent observations into topic files such as `topics/testing.md`; dream also runs on its own from time to time. xAI was explicit about what stays out: secrets, task state, tentative conclusions, and anything the repository or its docs already cover. Memory is on now for new sessions — `/new` or a fresh `grok` — and notes start after the first completed turn. This is a harness change, not another Grok 4.8/5 date. The agent that already edits a repo now has a place to put how that repo actually works.


