What the author of an agent can see

An agent someone else built is shaped by its author: the tables it keeps, the services it uses, where it sends results and who pays for it. Before you use such an agent, Donkit shows you in plain words what its author can see of your use — worked out from how the agent is actually set up, not written by the author.

Before you start

The question appears the first time you open an agent someone else built: from a published link, one shared with you personally, one published to your organization, or a team's page — and in your main chat when you pin one. It lists what applies to this agent, item by item, and offers two buttons:

• Agree and continue opens the agent.
• Leave closes the agent without using it.

Some people are never asked: the agent's owner, the Owners and Admins of the organization the agent belongs to, and people the owner gave Can edit. Built-in agents and your main assistant never ask either.

A published team asks once: one agreement covers the team and every agent it runs for you.

What the description covers

The description lists only what the author — and people on the author's side — can see of your use, and the places outside Donkit where what you share can go. Only the items that apply to the agent you are opening are shown. Together they can say:

• The agent's tables. Which tables the author can read in full, including your records. Tables whose records everyone who uses the agent can see. Tables where the records you add are visible to the author. When the agent belongs to an organization, that its Owners and Admins see the same records as the author. And which records stay in the agent if you leave it — see What stays. For more on how tables keep records apart, see Agent database.
• AI models the author connected. When your conversations are processed by AI models that the author or their organization connected instead of the ones Donkit provides.
• The author's accounts. When the agent works in the author's own mailbox, Google account or Telegram bot, the author can see there what the agent did for you.
• Outside services. The services outside Donkit that the author connected, where what you share with the agent can go. Each is named by its site, or called the author's own server when its address is not a public one.
• Automations. Where the author's automations send their results — a web address, email addresses at a domain, a Telegram chat — and that the author sees whether your runs of them worked.
• Your keys. If you create keys to use the agent from other programs, the author sees them in its list of keys, with your name and when each was last used.
• Who pays. When the author or their organization pays for your use, they see how much each person spends.
• The author's list of people. That the author sees you use the agent, with your name and email, and when and how often you use it — not what you write in your chats. Where the agent is open to the whole organization, its members see that you use it too.

On some self-hosted installations the agent's database does not keep records apart. There the description says that the author can see the records you add to every table.

Your email and profile

An agent someone else built always knows your name. Your email and the facts in your profile reach it only if you turn on Share my email and profile with this agent in the question — it is off unless you do.

To change your choice later, open My settings on the agent's page and go to Your data. Changing it does not ask the question again.

When the agent changes

The question comes back only when a change lets the author see more — a new outside service, a table the author can now read, a new place where results are delivered. The new items are marked New. A change that shows the author less does not ask again.

Until you agree again:

• the chat shows the question instead of an answer;
• your automations with the agent wait: your schedules are skipped but stay on, and deliveries to your webhooks and calls with your API keys are refused;
• you get one email the first time, Your automations with "Agent" are paused, with a link to review the agent.

Once you agree, everything picks up again. A run that had already started when the agent changed finishes as usual.

Leaving an agent and deleting your data

You can leave an agent someone else built and have your data in it deleted:

• on the agent's page, open My settings, go to Your data, and press Withdraw and delete my data, then Withdraw and delete; or
• in your account settings, open My data in agents: it lists every agent built by someone else that you agreed to use, with its state, and Withdraw and delete next to each.

At once:

• you leave the agent — for an agent shared with you personally you need a new invitation to come back; your membership in an organization is not touched;
• your conversations with it disappear;
• your schedules and webhooks on its workflows switch off;
• your API keys for it are deleted.

About 15 minutes later the deletion itself starts. It deletes your conversations, your files and the ones the agent made for you, what it remembers about you, your records in tables that keep records to their owner, your workflow runs, schedules and webhooks, your settings and connection choices for this agent, and your name and email in the author's list of people.

Until it finishes, My data in agents shows the deletion in progress, and you can't use the agent. Once it is done you can come back and agree again, like a new visitor.

Deleting your Donkit account does the same for every agent someone else built that you used. If the author removes you from the agent, nothing is deleted — you can still delete your data from My data in agents.

What stays

Leaving an agent does not delete:

• Billing records — what was spent and who paid. The author's usage pages keep the amounts.
• Your profile. It belongs to your account, not to the agent; manage it in Profile & memory.
• Records you added to tables shared with everyone. They have no owner, so they stay in the agent; the description names those tables before you agree.
• Your connections. They stay in your Connections library; only this agent's choice of them is removed.

In the apps

The current Android, iPhone, iPad and Mac apps ask the same question the first time you open an agent someone else built in the app — its chat, its app, a link or invite to it, or when you pin it. See Before you use someone else's agent in the apps.

Older versions of the apps show a message with a link to the agent's page on the web instead. Agree there, and the older app works with that agent too.

Leaving an agent and deleting your data in it is done on the web. To update the app, see Signing in from the mobile app.

For how the author shares an agent and what they see, see What visitors agree to; for API keys, see Agreement required before a key can call the agent.