---
title: Memory controls
description: The actions available in personakit review and what each one does.
status: pre-release
audience: people who run PersonaKit locally, contributors, and AI assistants
last_verified: 2026-11-12
---

# Memory controls

Run `personakit review`. Unconfirmed memories appear oldest first, with up to two supporting quotes.
A memory an assistant proposed is labelled as such, because its quote cannot be verified.

| Key | Action | Result |
|---|---|---|
| `a` | Approve | State becomes `approved`; served unlabelled |
| `r` | Restrict | State becomes `restricted`; never returned to assistants |
| `h` | Hide | State becomes `archived`; never returned |
| `f` | Forget | Deletes the memory and its quotes, after confirmation |
| `s` | Skip | No change |
| `q` | Quit | Stops review |

`--limit N` sets how many memories one session shows (default 10).

Decisions stick: a later sweep that finds the same memory again adds evidence but never undoes
your decision.

Not available in 0.1: editing a memory's wording, and changing a decision you already made.
`review` shows only unconfirmed memories, so an approved, restricted or hidden memory can be
forgotten with `personakit forget <memory-id>` but not moved to another state. Forget it and let a better version be extracted,
or restrict it.

Memories already read by an assistant stay in that conversation. A later restriction applies to future reads only.