Learn about me
Bring over what ChatGPT, Claude, Gemini, Grok or Copilot already knows about you, and choose line by line what your bots keep.
Two ways to do it
Two ways, same result: facts about you in your workspace memory, where every bot can recall them.
- Ask your Chief of Staff. Teach your Chief of Staff who you are walks through it as a conversation with Ember. Watch it if you would rather talk than click.
- Use the Learn about me screen. SettingsMemory has a section called Learn about me. It does the same job with a line-by-line review, so you read every line before it goes in. The rest of this page is about that screen.
Import with Learn about me
Step 1
Open Learn about me
Open SettingsMemory and find Learn about me. Under Where this came from, pick the assistant you are importing from: ChatGPT, Claude, Gemini, Grok, Copilot or another assistant. Murage labels every kept line with that name and the date, so you can always tell an imported fact from one a bot learned itself.
You're done when: you can see the prompt and a Copy the prompt button.
Step 2
Run the prompt in your other assistant
Choose Copy the prompt, open your other assistant, paste the prompt and send it.
The prompt asks for everything the assistant has saved about you as plain text under five headings: INSTRUCTIONS, IDENTITY, CAREER, PROJECTS and PREFERENCES. One item per line, each line starting with the date it was learned, or the word undated. It also tells the assistant to leave out bank, card and account numbers, passwords and other codes, because Murage does not keep them.
Read the answer before you copy it. If there is anything you do not want your bots to know, delete it now.
You're done when: you have the answer as plain text, under five headings.
Step 3
Paste it back and review
Paste the answer into Paste the answer here and choose Review what it found.
Murage reads the text with plain code, not a model, so nothing is sent anywhere at this point. The headings only decide which group a line is shown under: Instructions for every bot, About me, Projects and preferences, and Not under a heading for anything it could not place.
If it says Nothing to review in that answer, check that you pasted the answer to the prompt and not the prompt itself.
You're done when: you see the lines in groups, each with Keep and Skip.
Step 4
Keep the lines you want
For each line, read The words your bots will keep. Edit the words first if you like, then choose Keep or Skip. The Projects and preferences group also offers Keep all.
A kept line shows Kept with an Undo button. Undo removes it straight away. Later, you can forget any imported line in SettingsMemory like any other memory.
You're done when: the screen reads All reviewed. Your bots now know what you kept, and you can forget any line in Settings > Memory.
What the flags mean
Some lines carry a short flag. A flag asks you to read the line before you keep it. It does not stop you keeping it.
| Flag | What it means |
|---|---|
| Reads like an instruction | The line starts like a command, such as "always" or "never". Keeping it saves it as a note about you. It does not turn into a rule for your bots. For standing rules, use House rules. |
| Mentions the other assistant as you | The line says "you" about the other tool. Reword it so it reads as a fact about you. |
| Appears twice | The same words appear on an earlier line. Keep one. |
| The date did not look right, so it reads as undated | The line did not start with a usable date. It is kept as undated. |
| A number was masked | Murage found what looked like a bank, card or account number and hid it before showing you the line. |
| Over 1,000 characters, shorten it first | A single line can hold up to 1,000 characters. Edit it down, then keep it. |
Limits
- One paste. Up to 100,000 characters and 400 lines. A longer answer gets Paste it in parts, so split it and paste twice.
- One line. Up to 1,000 characters.
- Memory must be on. If it reads Turn memory on to keep lines, switch memory to Capture and recall or Capture only first.
- Notes, not orders. Imported lines are reference material about you. Keeping one saves it for recall. It does not change your bots' instructions or your capture settings.
- Bigger files. Notebook files and topic notes go in through Import existing notes on the same page, which takes larger files than a paste.
If you paste an export straight into a chat
Type a long, structured paste into a direct chat on the desktop app and Murage notices. Under your message it shows This reads like notes from another assistant. Murage left it out of memory. Review it to keep the lines you want. with a Review and keep button. That button opens Learn about me with your paste already filled in.
Murage does this so a pasted profile does not land in memory wholesale, with no chance to skip a line. The check looks for 12 or more lines of dated items or the five headings above. It is not used for code or logs. In a room, on the phone or from a messaging app you see the note without the button, so open the desktop app to review.
You can also just ask Ember in the chat to remember particular facts from what you pasted.
Your whole chat history
There is no one-click importer for a full data export, such as a ChatGPT conversations file. What works: put the file in a bot's working folder, or attach it in the chat, tell the bot what it is, and ask it to read it in parts and list the lasting facts, rules and preferences about you. Review that list, then ask it to remember the ones you want. Check them afterwards under Memory.
For most people the prompt is faster, because the other assistant does the searching.
No account sync
Murage never signs in to ChatGPT or Claude to read their memory, and it does not keep in step with them afterwards. You do the copy once, it takes a few minutes, and it comes with the review step above. From then on your bots' memory lives on your computer and does not depend on the other tool. A bot that signs in to its engine with your Claude or ChatGPT account uses that sign-in for the engine only. It does not reach the other tool's memory.
Passwords, cards and bank numbers
The prompt tells the other assistant to leave out passwords, bank, card and account numbers, ID numbers and other people's private details. If a number still gets through, Murage masks bank and card numbers before the line is shown to you, and flags it. Masking works on the common number formats, so still read the answer yourself before you paste it.
The same rule applies in chat: bots never save keys to memory, and when something they are about to remember could include a password or key, they ask you first.