> ## Documentation Index
> Fetch the complete documentation index at: https://docs.memloom.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Scopes

> Folders and tags: how memories and documents stay separated by area of life

Everything you save lands in exactly one **folder**. Folders are hierarchical
(`Personal life > People`), you create them, and every memory and every context document points
at one. **Tags** are free-form labels on top (`ai`, `twitch`, `girlfriend-moments`): a memory
has one folder and any number of tags.

The point is recall quality. Without scopes, a year of captured coding sessions and your life
memories compete in one ranking, and "plans with my sister" comes back with database trivia.
With scopes, a recall names the folder it is asking in and other folders stay out entirely.

## Two folders you never create

| Folder | What lives there |
| - | - |
| `uncategorized` | Saves that named no folder. The classification pass files them later; until then they are still recallable. |
| `global` | Cross-cutting preferences that should surface in every scope ("never use em dashes"). |

Both are created lazily and cannot be renamed or deleted.

## What a scoped recall searches

A recall with `scope: "personal-life"` searches that folder **and its subfolders**, plus
`uncategorized` (a memory awaiting filing must not vanish), plus `global`. It never searches
other folders. `scope: "*"` searches everything, which is the right call for deliberately
cross-scope questions. An unknown path is an error, not an empty result, so a typo cannot
masquerade as "you have no memories about that".

Browsing is stricter than recall: a folder listing (`memloom scopes`, the viewer sidebar,
`/memory/list?scope=`) shows just the subtree, with no widening.

## How things get their folder

* **You say so**: `--scope` on `save`, `recall` and `context add`; the `scope` parameter on the
  MCP tools; the viewer's move actions.
* **Capture maps mechanically**: a distilled Claude Code session lands in
  `projects/<name>`, derived from the session's working directory. No model involved.
* **The classifier files the rest**: saves that named no folder are classified against your
  folder tree (names and descriptions) during indexing. Below the confidence floor they stay in
  `uncategorized`. The classifier never invents folders; only you create them. It runs on
  its own, so it spends from the same daily budget as unattended capture and files at most 200
  memories per indexing run; a large backlog clears over several runs.

Folder descriptions matter: they are what the classifier reads. A folder described as
"streaming clips and Twitch moments" collects the right saves; an undescribed folder is just a
name.

## What never changes with scopes

* **A belief stays where it lives.** Restating a fact from another scope versions it in place;
  an exact duplicate merges into the existing folder. Moving a memory is always an explicit act
  (`memloom scopes set`, the viewer), and it moves the whole version chain.
* **Dedup and conflicts are owner-wide.** A contradiction between two scopes is still a
  contradiction.
* **Entities are global.** There is one `memloom` entity whether it shows up in work or in
  life; scope filters which *mentions* of it you see, never the entity itself.

## Addressing

Folders address by slug path: `Personal life > People` is `personal-life/people`, matched
case-insensitively, so renames that only change casing or spacing never break saved paths.
Renaming or moving a folder is a registry edit; nothing stored on memories rewrites.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.