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

# Memory management

> A persistent knowledge store Seb reads before every task — project memory and organization-wide conventions — so it doesn't relearn your codebase each session.

Memory is a persistent set of facts Seb reads before every task: conventions, decisions, hardware quirks, and toolchain preferences it would otherwise rediscover each session. It's injected into every agent — in the Platform, the CLI, and background jobs.

## Two tiers

| Tier                                     | Use for                                                                  | Where to find it                                                                            |
| ---------------------------------------- | ------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------- |
| **Project memory**<br />(this project)   | Errata, design decisions, and quirks specific to this board or codebase. | **Platform:** Project Settings → Memory<br />**CLI:** `/memory` to view, `/remember` to add |
| **Organization memory**<br />(whole org) | Team-wide conventions, toolchain preferences, how Seb should behave.     | **Platform:** Profile → Memory<br />**CLI:** `/memory` to view                              |

Both are read by every agent. Organization memory is shared with all org members, so keep it to genuinely org-wide facts.

## How facts get saved

* **Seb saves project memory automatically** as it works — when it learns something durable, it writes it to project memory.
* **Organization memory needs your approval.** When Seb proposes an org-wide fact, a confirmation chip appears in chat; nothing is saved until you approve it.
* **You can add facts yourself** — in the CLI with `/remember <fact>` (project memory), or by editing organization memory directly in **Profile → Memory**.

View what's stored with `/memory` in the CLI (org memory + project memory), or in the Platform — project memory under **Project Settings → Memory**, organization memory under **Profile → Memory**.

## Isolated projects

Project owners can mark a project **memory-isolated** in project settings. On an isolated project, Seb's project memory still works, but nothing is ever promoted to organization memory — use it for NDA-bound, export-controlled (ITAR/EAR), or per-client work where facts must not leak across the org.

## Guardrails

* **Secrets are refused.** Writes containing key material, private keys, tokens, or verbatim code/file contents are rejected — keep memory at the level of facts and conventions, not data.
* **Size cap.** Each store is capped (\~32k characters); prune stale entries before adding more.
* **Audited.** Every write is logged with who wrote it and when, for org oversight.

## See also

* [Slash commands](/reference/slash-commands) — `/remember`, `/memory`
* [Guide → 3. Run agents](/guide/run-agents)