---
title: Memory
description: Facts and preferences a Space can remember across Jobs. Memory is separate from chat history and model choice.
updated: 2026-09-12
canonical: https://turnpanel.com/docs/memory
markdown: https://turnpanel.com/docs/memory.md
llms: https://turnpanel.com/llms.txt
---

# Memory

Facts and preferences a Space can remember across Jobs. Memory is separate from chat history and model choice.

## Two layers

- Core memory is the short list of important facts and preferences shown directly to the agent.

### Important facts shown directly to the agent

**Context:** Core memory

The Core memory view shows the compact facts given directly to the agent.

- **Core memory** [active]: Stable facts
- **Preference · constraint · relationship · project** [ready]: Durable context categories
- Episodic memory stores more facts and events. The agent searches it when needed, and entries keep a link to where they came from.

### Searchable memories with their sources

**Context:** Episodic memory

The Episodic memory view shows searchable entries and their sources.

- **Episodic memory** [active]: Searchable facts and events
- **Status · access · source chat** [ready]: Entry metadata
- **Archived** [ready]: Separate dedicated route
- Past Job conversations are stored separately from memory.


## Recall and learning

When a Job starts a turn, TurnPanel provides a small set of memories that seem relevant. After work finishes, background maintenance may save useful facts in that Job’s Space. It never moves private Space facts into General on its own.


## Correcting memory

Tell the agent exactly what to remember or forget, or open Memory to review and change entries. If an old fact still appears, check both Core and Episodic memory. Never store passwords or keys in memory.


## Limits

Memory search looks for similar meaning, so it can miss a fact or return the wrong one. Treat memory as helpful context, not proof. Stored memory can stay on your computer even when an online model answers the Job.