Skip to content

Profile & memory

Agents remember things about you in two places: a short profile that your own agents keep in mind, and a memory that each agent keeps about you and your work together. Both belong to you, and you can see and change all of it.

Everything here is kept per workspace: your personal account and your membership in an organization are two separate personas, each with its own profile and its own memory.

Your profile

Your profile is a short list of facts in the form name: value — your role, your company, how you like to be addressed, the formats and languages you want, standing preferences. It holds up to 30 facts. Your name is not part of it: agents already know it from your account.

  • Always in view for your own agents. Donkit, the agents of your own workspace and the built-in agents see your profile on every message. An agent someone else built — including one published to your organization, unless you are an Owner or Admin there — sees it only if you chose Share my email and profile with this agent; see Your email and profile.
  • Agents keep it current. When you tell an agent something like "I'm the head of procurement" or "send reports as PDF", it adds or updates the fact itself. When the profile is full, it has to merge or remove a fact first. If your profile shows where each fact came from and a Remove button next to it (see Where a fact came from), agents cannot remove a fact: when the profile is full, they ask you to remove one.
  • Only agents of your own workspace change it. Donkit, the agents you or your organization built and the built-in agents can add or update a fact — and only while you are the one chatting with them. An agent someone else shared with you never changes it, even when you share your profile with it. Calls through an API key, workflow runs and agents calling other agents never change it either.
  • You have the last word. Add, edit or delete any fact yourself.

Where a fact came from

If your profile shows where each fact came from and a Remove button next to it, agents save what they learn about you at once, without a confirmation card. The line under each fact on Account → Profile says where it came from: added by you, said by you in a chat, a public page, or your own mail (your sent letters and signature), with a short note saying where.

  • Remove takes a fact out at once. Agents stop seeing it, and they are refused when they try to save it again, also when you say it again in a chat. The fact moves to Removed.
  • Return brings it back as it was; Delete for good deletes it. Adding the same fact again yourself, or keeping it on an earlier facts card, brings it back too. Returning a fact needs a free place among the 30.
  • The last 30 facts you removed are kept; older ones are deleted for good. Apart from them, the last 30 facts put away by Don't look are kept too, so pressing it never deletes a fact you removed.
  • Don't look in Direct puts away every fact from a public page: they move to Removed with the reason. You may look there brings back exactly those while there is room among the 30; what does not fit stays in Removed with its reason and Return. A fact you removed yourself never comes back this way. Return of a put-away fact works only while the look-up is allowed with You may look.
  • Your role, company, people, key document, writing style and signature come first, and the signature is edited in place.
  • What you wrote or said yourself is replaced only by what you say again, never by what an agent found. The account assistant's changes show as saved by an agent.
  • A fact you edit becomes yours: it shows as added by you, and the note on where it came from goes away.
  • Only you remove or return a fact. The account assistant cannot.
  • What Donkit reads in mail it sorts in the background, or a task reads on its own, never becomes a fact about you.
  • A ready errand card in Direct names the facts it relied on and offers to rewrite the errand without one that has left memory.

Where to find it:

  • Web — Account → Profile.
  • Android, iPhone and iPad apps — Settings → Personal profile; Mac app — Settings → Account → Personal profile. The counter at the top shows how many of the 30 places are used. In the apps, deleting a fact deletes it for good for now.

Who can see your profile

An agent someone else built sees your profile only if you chose to share your email and profile with it. Once you do, the agent reads your profile on every message, though its owner still cannot open your conversations with it. Keep passwords, card numbers and other secrets out of your profile.

You don't need to tell agents your time zone: every message carries your device's time zone, so "tomorrow at 9" means 9 in the morning where you are.

Your memory with each agent

Each agent keeps its own notes about you and your work together — tasks you gave it, decisions you made, results you reached ("picked hotel X for the Milan trip"). This memory is kept per agent and per person: other people who use the same agent never see your notes, and one agent's notes about you don't leak into another agent.

To see what an agent remembers:

  • Web — on a published agent's page, open My settings → Memory; in your chat with Donkit, choose My memory.
  • Mobile apps — tap the memory button in the chat header (My memory with this agent).

The list shows the newest notes first; Load more brings older ones. You can add a note, edit one in place or delete it. An edited note stays where it was in the agent's memory; a deleted one is forgotten.

What agents learn for everyone

Agents that people build also have agent learnings — lessons the agent draws from its own work that help everyone who uses it, such as "invoices from this supplier always need a PO number". The agent adds them rarely and on its own, and only in a conversation with its owner, an Owner or Admin of the owner's organization, or someone who can edit the agent; in anyone else's conversation the save is refused. It declines requests to remember something "for everyone", and never puts anything about one specific person there. Your own notes and your profile never go into them. The agent's owner reviews and edits them — see Agent learnings & agent files.

Connections and the account assistant in the Android app

Settings → Connections opens your Connections library in the app — the MCP servers, Google accounts, mailboxes and Telegram chats you've connected, grouped by kind, each with its status. Every row has one main button: Sign in or Reconnect when access is needed, otherwise Test. The row's menu lets you:

  • Rename the connection;
  • Update secrets — new keys for an MCP server, a new password for a mailbox or a new bot token for Telegram;
  • see which agents use it (Used by agents) and Unlink it from one of them — the connection itself stays in your library;
  • Delete it — the confirmation names the agents that still use it.

The Code hosting group lists your GitHub and GitLab rows. A row that is not connected offers to sign in with the provider, or with a personal access token; a connected one shows the account and lets you reconnect it when access is lost or disconnect it. The same group is in the iPhone, iPad and Mac apps. You can also connect GitHub or GitLab straight from the Web developer, without opening Settings first.

The + button at the top adds an MCP server, a Google account, a mailbox (checked with the mail server before it is saved) or a Telegram chat — or hands the job to the assistant with With the assistant.

The account assistant works in the app too. A round button at the bottom-right of the Settings, Profile, Personal profile, Devices and Connections screens opens it; its header has New chat and History. It is a chat only — no attachments, canvas, files or model picker. When it needs a password, token or API key it opens a secure form right in the chat, and when a service needs you to sign in it shows a button for that. Links it gives you to account pages open inside the app.