Persistent memory features introduced for Grok builds

Grok builds now feature persistent memory capabilities designed to retain important project details and user preferences across different working sessions. This system aims to reduce the need to repeatedly explain project context by keeping track of specific information that matters most for future tasks.

What the system remembers

The memory captures details most likely to matter in later sessions. This includes how a team writes and reviews code, key decisions along with their underlying reasoning, and durable facts about the project. Durable facts range from the location of a subsystem to the specific commands required to run a test suite.

However, certain types of information are deliberately excluded. The system leaves out task state, tentative conclusions, secrets, and anything that is already covered within the repository or its documentation.

Notes are divided into two categories. The system maintains project-specific notes as well as a global set of preferences that apply everywhere.

Background capture and organization

Note capture occurs in the background after a conversation turn completes. This process never blocks the active session. Grok reviews the finished turn, writes anything worth keeping as a markdown note, and then continues.

Over time, a command called /dream folds these individual notes into topic files, with one file per subject. As a result, each project ends up with a small set of organized references. When returning to a project, Grok reads the specific topic files that cover the area it is about to work in. Instructions provided in the current conversation always take precedence over anything stored in a note.

Browsing and editing notes

A command called /memory opens a read-only browser displaying every memory file. Files are grouped by scope and include a preview of the selected file. This interface provides a fast way to review what a session produced and locate any file that requires editing if a note is incorrect.