> ## Documentation Index > Fetch the complete documentation index at: https://wonder.design/docs/llms.txt > Use this file to discover all available pages before exploring further. # Memory and design preferences > See and edit what your agents remember, the design preferences they apply to every design, and the instructions every chat loads. Your agents carry what they learn from one chat to the next in two ways. **Agent memory** is what Claude Code and Codex remember about your repositories and about you. **Design preferences** are the design rules you've asked agents to remember, such as "use 8px corner radius on cards", which they apply in every chat. You see and change both in **Settings > Memory**. **Before you start** · Desktop app · Free plan Everything here stays on this computer. Agent memory is kept in your agents' own folders and is shared by everyone who uses this computer. Design preferences are kept for your Wonder account on this computer: someone else who signs in to Wonder here has their own, and they don't follow you to another computer. ## Save a design preference Ask an agent in a chat, in your own words: "Remember that I always want 8px corner radius on cards." The agent saves it, and the chat shows **Saved to your design preferences**. From then on, every chat on this computer is told it. * Agents save a preference only when you state one or ask them to remember it. They never save one they inferred from your files, your designs or another agent. * Depending on the chat's mode, the agent may first ask **Save this to your design preferences?** Approve it to save. * Each preference is one line of up to 280 characters, and you can keep up to 50. * Project conventions, such as which components a repository uses, belong in the repository's instruction files instead. See [Instructions every chat loads](#instructions-every-chat-loads). Instructions in the chat and the repository win over a design preference. ## Edit or delete a design preference Open **Settings > Memory**. **Design preferences** lists each one, with who saved it last and when, such as **Edited by you** or **Saved by an agent in** the chat it came from. ![Design preferences in Settings > Memory with one preference, Use 8px corner radius on cards, saved by an agent in a chat, and its edit and delete buttons](https://mintcdn.com/wonder-0722b0f6/d_fx9trSoNCyzqAi/images/control-your-agent/design-preferences.webp?fit=max\&auto=format\&n=d_fx9trSoNCyzqAi\&q=85\&s=8e5c8a6fca7ed9067a45ce112b05ad3d) * **Edit:** click the pencil next to a preference, change the wording under **Preference** and click **Save**. Agents apply the new wording from their next message. * **Delete:** click the trash icon and confirm with **Delete preference**. Agents stop applying it in every chat on this computer from their next message. This can't be undone. You can also ask an agent in a chat to reword or forget a preference. ## See what your agents remember **Settings > Memory** shows what your agents remember about you: * **Quick chats**: what each agent remembers for chats that aren't in a repository. * **About you**: what your agents remember about you in every repository. Codex keeps one memory for all your repositories, and it shows here. To see what agents remember about one repository, open the repository's menu in the sidebar, choose **Repository settings** and open **Memory**. Claude Code keeps a memory per repository, used by every copy of it on this computer, including its worktrees. Each agent's row shows whether its memory is **On** or off, and where it keeps it. Click **Show in Finder** to open that folder. ## Edit or delete what an agent remembers * **Claude Code:** click **Open in editor** on a note to change it in your editor. To delete one, click **Delete** and confirm with **Delete note**. The note's file and its line in the memory index are removed, so later chats in that repository no longer know it. This can't be undone. * **Codex:** **What Codex knows about you** is the summary Codex gives every chat. Click **Open** to read it in your editor. Codex writes its memory itself, so Wonder can't delete from it: click **Ask Codex to forget** next to an entry instead. ## Ask an agent to forget something * **Codex:** click **Ask Codex to forget** next to the entry. Wonder opens a new quick chat with the request written for you. Read it and send it. The chat shows **Codex will forget this**, and Codex applies it to its memories later. * **Claude Code:** delete the note in **Settings > Memory** or in the repository's **Memory** settings, or ask Claude in a chat to forget it. * **Design preferences:** delete the preference in **Settings > Memory**, or ask an agent in a chat to forget it. The chat shows **Removed from your design preferences**. ## Turn memory on or off Wonder shows each agent's memory but doesn't switch it on or off. Each agent decides that in its own settings: * **Claude Code** keeps memory on unless you've turned it off. Run `/memory` in Claude Code to turn it on or off. * **Codex:** run `/memories` in Codex to turn memories on, or set `features.memories = true` in its config file. In the EEA, the UK and Switzerland, Codex offers memories only after you opt in. When an agent's memory is off, its row says why, for example **Off until you turn on** the setting it needs. ## Share memory between your agents **Share memory between your agents**, in **Settings > Agents > Access**, tells each chat what your other agents remember. A Claude Code chat then also knows what Codex remembers and what your personal Codex instructions say, and a Codex chat knows Claude Code's. It's on by default and shows once Claude Code or Codex is signed in. When you turn it off, chats that already got the shared memory are told to disregard it. The change applies when a chat's agent next starts. ## Instructions every chat loads **Instructions your chats load**, in **Settings > Memory**, lists the files in your home folder that agents read when a chat starts: * Claude Code: `~/.claude/CLAUDE.md`, and the rules in `~/.claude/rules/`. * Codex: `~/.codex/AGENTS.md`, or `~/.codex/AGENTS.override.md` when you have one. Click **Open in editor** to change a file. Each row says which agents read it. With **Share memory between your agents** on, Wonder also gives each file to the agent that doesn't read it itself. If you don't have any yet, add one of these files to give every chat standing instructions. A repository can have its own instructions too, in an `AGENTS.md` or `CLAUDE.md` file. The repository's **Memory** settings list them. ## If something goes wrong * **"Couldn't load saved preferences."** Wonder couldn't read your design preferences on this computer. Open **Settings > Memory** again, and restart Wonder if it keeps happening. * **"You already have a design preference that says this."** The same preference is saved already. Edit the existing one instead. * **"You have reached the limit of saved design preferences."** You have 50. Delete one you no longer need. * **"Update Wonder to see what your agents remember."** This version of the desktop app can't read agent memory. Update Wonder. * **"That file is no longer there."** The agent removed or moved the note since the list was shown. Open **Settings > Memory** again to see what's there now. * **An agent doesn't seem to remember anything.** Check its row in **Settings > Memory**: if it says **Off**, turn memory on in that agent as described in [Turn memory on or off](#turn-memory-on-or-off). ## Related * [Agent and chat preferences](https://wonder.design/docs/agent-and-chat-preferences.md) * [Add to chat from the canvas](https://wonder.design/docs/add-to-chat-from-the-canvas.md) * [Design with AI](https://wonder.design/docs/design-with-ai.md) This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.