we_are_coded.by CODE · The world, decoded
БГ
SpaceXAI

Grok Build now remembers your project between sessions and writes its own notes

SpaceXAI News · event date: 16 September 2026Builders

After every completed turn the agent records conventions, decisions and project facts as markdown notes. Secrets, tentative conclusions and anything already documented stay out of memory, and an instruction in the conversation beats a note.

In short
  • Notes are kept per project, plus a global set for preferences that apply everywhere.
  • /dream folds notes into topic files, /memory shows them read-only.
  • Memory applies to new sessions started after the update.
Checked on1 October 2026Responsible editorTsvetelin IvanovHow we workMethod · Corrections

“Run the tests with just test, not with cargo test.” You say it once. Then again the next day. And the day after that.

Grok Build so far was like a new colleague who comes in every morning with an empty head. Smart, fast and with no memory of yesterday at all.

The facts: on 16 September 2026 SpaceXAI added memory to Grok Build. After every completed turn Grok reviews in the background what happened and writes the durable things down as a markdown note: how the team writes and reviews code, decisions and the reasons behind them, facts about the project - from where a module lives to which command runs the tests. Task state, tentative conclusions, secrets and anything the repository or the docs already cover are not kept. Notes are kept per project, plus a global set for preferences that apply everywhere. The /dream command folds notes into topic files, and /memory opens a read-only browser of all memory files. When you return to a project, Grok reads the topics that cover the area it is about to work in; instructions in the current conversation take precedence over notes. Memory applies to new sessions.

Two things are done well and deserve to be said. Secrets stay out: the key, the database password and the token from the environment are kept out of the notes by rule. And the conversation beats memory, so an old note cannot shout down what you are saying now.

A good memory also knows what to forget.

The third one is subtler. The notes are ordinary markdown files and you can open them. When the agent remembers something wrong, and it will, you know where to fix it. Memory you cannot see is more dangerous than no memory at all.

In a team it is worth opening /memory at the end of the first week and reading what the agent has decided it knows about you. You will learn something about the agent, and about the team.

The visual is generated code art. No third-party images.
Follow usFacebookLinkedIn
Official primary sources
→SpaceXAI News - Memory in Grok Build, 16.09.2026
Original: https://wearecoded.com/en/articles/grok-build-pamet-mezhdu-sesiite.html
ShareFacebookXLinkedInTelegramWhatsApp
← Back to all news