# Workspace memory

> What Koris remembers: session summaries, Markdown topics, and optional company website learning.

Koris remembers what matters about your workspace in two ways: a short summary that carries each conversation into the next session, and a set of Markdown notes about your projects, preferences, procedures and lessons learned.

## When Koris writes memory

Memory is written at two moments only:

- **When a conversation is summarized.** A conversation that gets long, or that you return to after about five hours, moves to a fresh session. In the background, Koris summarizes the old session (usually within seconds) and adds the summary to the new one, and merges any lasting facts into the notes. You can keep chatting while it does; the chat shows when a summary is in progress.
- **When Koris learns your company website** (below).

Ordinary messages, chats with specialists and group work don't write to memory on their own.

## What the notes look like

The notes are organized in four folders:

| Folder | What goes there |
| --- | --- |
| Projects | One note per project you work on |
| Preferences | How you like to communicate, and the tools and stack you use |
| Procedures | How things are done in your workspace |
| Lessons | What was learned along the way |

Each folder has an index listing its notes with a one-line description, and there is a main index for the whole workspace. Notes appear as they're needed; a new workspace starts empty. Earlier versions of a changed note are kept, so an update never loses what was there.

## How agents use memory

In conversations with memory on, the Chief sees the main index, recent summaries, and the notes most related to what you just asked. It can also open any note or folder index to read it in full. Agents read memory; they never edit notes directly.

Groups and work delegated to specialists don't use workspace memory.

You can browse the notes yourself in the Chief's **Memory** tab.

## Company website learning

When you create a workspace, you can give your company's website (a bare domain such as `example.com` works). If you don't, the Chief asks once during onboarding; you can skip it.

Koris then reads a dozen pages of the site in the background, respecting the site's robots rules, and saves what it learns about your company to the Projects and Preferences notes. Website content is treated as information, never as instructions to the agents. Only public pages are read, and pages that need JavaScript to show their text are not supported.

The notes appear in the Memory tab when the learning finishes. Changing the website later in settings doesn't start a new learning run on its own.