Mythex Now Remembers Who You Are and How You Work
Memory, About you, project and workspace knowledge, and past-chat search: Mythex stops starting from zero in every chat. See, edit or turn off all of it.
Mythex Team · · 4 min read
Mythex no longer starts from zero every time you open a chat. The agent now remembers short facts about you across all your projects, follows standing notes you set for a project or a whole workspace, and can search your earlier chats when you refer back to them. You can see, edit, delete or switch off all of it in Settings → Memory.
Why we built it
Before this, every new project began with the same briefing: what your business is, who it's for, that you're not technical, that prices are in your local currency, that you prefer short answers. People repeated themselves, and the agent still missed things they had said a week earlier in another project.
The big AI assistants have converged on a similar answer to this problem: a small set of remembered facts, some standing instructions you write yourself, and a way to look back through past conversations. We built the same three pieces for Mythex, with one difference that matters for a building tool: rules can belong to a project or a workspace, and be shared with everyone who builds there.
The three parts
| What it is | Who writes it | |
|---|---|---|
| Memory | Short facts about you, like "Runs a bakery" or "Prefers short answers" | Learned from your chats, or added by you |
| About you | A few lines in your own words about who you are and how you like to work | You |
| Knowledge | Standing notes for a project or workspace, like "Never change the checkout" | You and your team |
Memory and About you belong to you and follow you into every project. Knowledge belongs to a project or workspace.
How the agent uses it
At the start of every message, the agent is given your About you, your memories, and the knowledge for the project you're in. It uses them without being asked: it replies in your language, matches your taste, and doesn't ask for things you've already told it.
There's a difference in weight:
- About you and knowledge are instructions. The agent follows them.
- Memories are background. The agent uses what helps, and what you say now always wins over something it remembered.
It won't recite your memories back unless you ask. Ask "What do you know about me?" in any chat to see.
How memories are made
It learns from what you write. When you mention something lasting — your business, your audience, your brand colours, how technical you are — Mythex saves it as a short fact. This happens in the background after you send a message, so it never slows down a reply.
Only your own words count. Nothing is learned from files, websites, tool results or the agent's replies. Requests about the task at hand ("add a login page") aren't remembered either; those belong to the project, not to you.
You can ask directly. Say "remember that…", "from now on…" or "forget that…" in any chat.
You can add them yourself in Settings → Memory.
Searching past chats
Memories are a short summary, not a transcript. When you say "use the colours from last time", the agent can search your earlier chats — in this project or across all of them — and answer from what it finds. It only searches chats you can already open yourself, and it's told to check before saying it doesn't remember something.
How to set it up
- Open Account settings → Memory → Manage, or go straight to Settings → Memory.
- Write a few lines in About you, for example: "I run a small bakery. Keep things simple, I'm not technical."
- Check the three switches: Use memory (the master switch), Learn from my chats, and Search past chats.
- For rules about one app, open the project, then Project settings → Knowledge and write what the agent should always know: audience, brand, things to leave alone.
- For rules across every project, the workspace owner can fill in Workspace knowledge — company name, brand colours, fonts. If project and workspace knowledge disagree, the project wins.
Changes apply from your next message.
Teams and privacy
We designed this so it's clear whose memory is whose, and so nothing is hidden from you.
- Your memory is yours. When a teammate chats in a shared project, the agent uses their memories, never yours, and nothing learned from your messages is saved to anyone else.
- Knowledge is shared with everyone on the project or in the workspace. Only the workspace owner can change workspace knowledge.
- You can see everything. What the agent carries about you is exactly what's on the Settings → Memory page, plus your project knowledge. Each memory shows where it came from and links to the chat it was learned in.
- No secrets. Anything that looks like a password, API key or token is refused. Keys belong in project secrets.
- Sensitive topics are left out — health, religion, politics and the like, exact addresses and ID numbers — unless you explicitly ask.
- Off means off. With Use memory off, the agent reads nothing about you and saves nothing. Project and workspace knowledge still apply.
Limits
- About you holds up to 3,000 characters; project and workspace knowledge up to 10,000 each.
- Each memory is one short fact, up to 300 characters, and you can keep up to 200. When the list is full, the oldest learned memory makes room. Ones you typed or asked for are never removed without you.
- Deleting a chat doesn't delete memories learned from it, and deleting a memory doesn't delete the chat. Past-chat search can still find something you said in a chat you kept.
The full reference is in the docs: Memory and knowledge. If you want to get more out of each message, our guide to writing prompts for AI app builders pairs well with a good About you.