Memory

nod remembers lasting facts and preferences about you, so you don't have to repeat yourself.

You can see everything nod remembers, where it learned it, and change or forget it in Settings → Memory.

What nod remembers

Kind	For example
About you	Where you work, your city, your languages
Preferences	Window seats, short replies, metric units
People	Who your manager is, your partner's birthday
Routines	You review invoices on Mondays
Lessons	Never send invoices from your personal email

Where nod learns from

Notes come only from your own messages and what you write or paste in Settings → Memory. nod never takes a note from its own replies, web pages, emails or tool results. Text from the web gets in only if you paste it yourself, into a short message or Settings → Memory.

When you tell it. Say "remember that I'm vegetarian" or state a lasting fact, and nod saves it as a short note, such as Diet: vegetarian. It can be something you said a little earlier: "remember that address" saves the address from your last few messages. Each note links to the message it came from, and nod is told to keep your names and numbers exactly as you wrote them. Messages longer than 4,000 characters can't be saved this way; add a shorter note yourself.
In the background. About once an hour, nod looks through your recent messages for what you said about your life: people, places, plans, things you bought, events with their dates and what you like. It skips messages with attachments and messages over 4,000 characters.

nod saves silently. It doesn't post a notice in chat.

Skills nod learns

After a task that took several steps or a workaround, nod writes down how it did it: the steps on that site or app, the gotchas and what to check at the end. The next time the same job comes up, it follows its own skill instead of working it out again. When a site changes, nod fixes the skill. nod saves skills only while Save new notes is on, and not for a task you ask it not to remember.

Before a skill is saved, the memory model replaces names, addresses, contact details, order or account numbers, amounts and message content with placeholders. It's a model, so it can miss something; every skill is under Skills for you to check. Because nod writes them from what it did and read, a web page can shape a skill. A skill never lets nod act: sending, paying or deleting still goes through the trust check.

How a memory grows

Something nod picks up from one message is Mentioned once: nod uses it, but as less certain. It becomes Learned when a second message repeats it.

nod also searches your earlier messages in the current conversation when you ask about something you told it, so it can answer from the conversation itself, with the date you said it. When you send a message, it also looks up your earlier messages that match it, so a recommendation can use what you told it weeks ago. After you reset the conversation, the old one isn't searched.

stateDiagram-v2
  state "Mentioned once" as once
  state "Let go" as gone
  [*] --> once: heard once
  once --> Learned: said again
  once --> Confirmed: you confirm
  Learned --> Confirmed: you confirm or edit
  once --> gone: 60 days quiet
  Learned --> gone: replaced or expired
  Confirmed --> gone: replaced or expired
  gone --> Confirmed: you bring it back

Replaced and expired notes, and guesses not mentioned again for 60 days, move to Let go instead of being deleted. When memory reaches its size limit, the oldest notes in Let go drop off first. To delete a note right away, select Forget or ask nod to clear its memory.

Sensitive topics

nod doesn't learn about health, religion, politics, sexuality, gender identity, ethnicity, immigration status or criminal history unless you ask it to. To let nod learn them, turn on Remember sensitive topics in Settings → Memory.

Your writing style from Gmail

When you connect Gmail, nod reads your own text from recent sent emails. It removes quoted, forwarded and automated mail first. It saves only how you write, with no facts, names or quotes, as one Preferences note labelled Learned from your sent email. A newer version replaces the old one.

Bring memory from another assistant

In Settings → Memory, select Import, then Copy prompt.
Paste the prompt into ChatGPT, Claude, Gemini or any other assistant, and copy its answer.
Paste the answer into nod. nod lists the notes it found under About you, Preferences, People and Routines.
Remove anything you don't want nod to keep, then select Import.

Import memory with the notes found in a pasted answer, grouped under About you and Preferences, above Back and Import 14 notes.

Each note is saved as written and marked Confirmed. nod skips notes you already have and lines that look like a password, code or card number.

Imported notes are like notes you add yourself: they help nod understand you and never let it act.

Review and change memory

Open Settings → Memory. Filter by All, Mentioned once, Learned, Confirmed, Let go or Skills, or search. Notes from chat show the messages they came from, and skills show the task they came from.

To	Do this
Fix a note	Select Edit
Tell nod a guess is right	Select Confirm
Keep a note only for a while	Select End date
Remove a note for good	Select Forget
Fix or remove a skill	Open it under Skills, then Edit or Forget
Restore an older note	Open it under Let go, then Bring back
Add something yourself	Select Add note
Bring notes from ChatGPT	Select Import
Download everything	Select Export for a Markdown file
Stop saving new notes	Turn off Save new notes

You can also tell nod in chat, for example "forget my old address" or "stop remembering things".

Check memory from Telegram or WhatsApp

Send /memory (or /context) to get a short summary. It doesn't use the model or cost anything. Telegram also receives the full export as a document.