Assistant

The dashboard has an assistant. It sits in a panel beside the page you are on, and you ask it about your workspace in plain words. It can tell you why an invocation failed, which of your Functions fail most, or what a Function’s code does. It can also make changes, and each change waits for you to approve it.

It acts as you. It sees what your role in the workspace lets you see, and nothing more.

What it can do

It reads your Apps, your Functions and their files, their versions, their invocations and their environment variables. It never reads a secret’s value. With your workspace’s permission, it also reads the logs your Functions printed. Reading logs says why that needs a yes first.

It can change things too:

• deploy a Function, roll one back, run one, or fetch a path from one;
• attach a domain to an App, or cancel a domain claim;
• change a Function’s files in a draft, show you a preview, and then promote the draft;
• ask you for an environment variable or a secret, which you type on a card in the conversation;
• ask you for files, such as a logo or a CSV, which you add on a card.

It can’t export your workspace, allow outbound calls or take a domain off an App. When a fix needs one of these, it tells you where in the dashboard you make the change.

It can’t install npm packages, and no build runs when it deploys. So it can’t deploy a template that needs a build. That is every template except landing-page. For those it gives you two commands to run on your own machine, which needs Node.js: npx wawesome init --template <name>, then npx wawesome deploy from the new directory.

Each message carries the page you sent it from, so the assistant knows which App, Function or invocation you mean. “Why did this fail?” is enough.

Approval cards

A change never runs straight away. The assistant says in one line what it wants to do and why, and a card appears in the conversation. You press Approve or Deny. When you deny it, you can add a note, and the assistant reads it.

An unanswered card expires after an hour, as if you had denied it. So does a card you leave waiting when you send a new message.

On most cards there is a third button, Always allow for this App. It approves this change, and every later change of the same kind to the same App, in this conversation only. A new conversation starts with no rules. The rules you have made are listed above the message box, and you can revoke any of them there.

Promoting a draft always asks. There is no Always allow for it, because a promote puts new files live. Before you promote, the card shows the files the draft changed and the lines changed in each, against the version the draft started from. If the draft changes after the card appears, the card shows no changes and the promote is refused.

While the assistant changes a draft, each edit shows in the conversation with the number of files it changed and the lines it added and removed. Nothing there waits for you, since a draft is not live. Open an edit to see the lines that one edit changed.

Every change you approve is recorded in your workspace’s action record as your act. The row also says it came from the assistant.

Sending a message while it works

You can send a message while the assistant answers, for example “use the other App” or “don’t deploy yet”. It shows as Queued until the assistant reads it. The assistant reads it after the step it is on, so a change that is already running finishes first. Then it goes on with your message in mind, in the same answer.

If the answer ends before the assistant reads your message, for example because you stopped it, the text goes back into the message box, so you can send it again.

Stopping an answer

While the assistant answers and the message box is empty, the send button becomes Stop. Press it and the answer ends within a second or two, keeping what it has written so far. A change that is already running finishes first, so a deploy is never cut off halfway. The assistant still knows about that change in your next message. To ask the same message again, press Retry under the stopped answer.

Turning it on and off

An Owner turns the assistant on or off for the whole workspace, in Settings, on the AI tab. When it is off, nobody in the workspace sees the panel, and nothing from the workspace goes to the model provider.

A workspace created now starts with the assistant on. A workspace created before the assistant existed starts with it off, and stays off until an Owner turns it on.

Turning it off deletes every conversation in the workspace. They can’t be brought back.

On the same tab, an Owner decides whether the assistant may make changes at all. With that off, it only reads. Everybody in the workspace can see both settings.

Reading logs

Logs can hold your own users’ data, such as an email address a Function printed. When the assistant reads a log, that text goes to the model provider. So reading logs is off until an Owner or an Admin allows it, once, for the whole workspace.

The first time the assistant needs logs, the conversation asks:

• An Owner or an Admin sees a card. Allow turns log reading on for the workspace, and the assistant reads the logs and goes on. Deny leaves it off, and the assistant goes on without them.
• Anyone else is told that an Owner or an Admin can allow it in settings. The assistant goes on without the logs.

An Owner or an Admin can turn it on or off later, on the AI tab in Settings. Each change is in the workspace’s action record.

This setting is for the dashboard assistant only. An agent you connect over MCP reads logs with the credential you gave it, as before.

What we keep

A conversation is kept for 30 days after its last message, then deleted.

What a tool answered, for example a page of logs, is cleared from the conversation after 14 days, when the logs it could quote expire anyway. Your messages and the assistant’s answers stay until the conversation goes.

A conversation is yours. Other people in the workspace can’t read it.

Don’t write passwords, keys or other secrets in a message. What you write is stored with the conversation and sent to the model provider. When the assistant needs a secret, it asks for it on a card, and it never sees the value. The Privacy Policy says what goes to the model provider and what it keeps.

How much you can use

Each plan includes a monthly budget for the assistant. A model step costs a little, and a long answer takes several steps, so the budget counts what the steps cost and not how many messages you send. Limits has the budget for each plan.

The dashboard shows the budget as a share, not an amount. From 80% it tells you what share of the month is used and when it resets. It resets on the 1st of the month, at 00:00 UTC.

When the budget is used up, the assistant stops and says when it comes back. A daily limit applies on every plan too, whatever is left of the month. It comes back at 00:00 UTC.

One answer can take a limited number of steps and tokens. When it reaches either, the assistant stops and says so. Send another message to go on.