AI on the platform

AI shows up in two places, at the two ends of the same document:

Where	What it does	Page
While preparing	Places signatures and fields from a sentence you write	Prepare a document with AI
While signing	Answers questions about the document in front of you	Ask questions about a document

They are different features for different people — one serves the sender, the other the recipient. What they share is everything else on this page.

One balance, counted in tokens

Both spend from the same organisation balance. There is no quota for preparing and another for reviewing: there is one, and every request debits it.

The unit is the token. The application explains it like this:

“Tokens” are the units the AI model uses to read the document and generate its reply.

Two orders of magnitude worth knowing:

A typical document review runs to about 15,000 tokens.
Every message reserves a minimum of 1,000 tokens.

The real figure depends on the document’s size and complexity, and it is the AI provider that reports it — not an estimate by the platform.

The review panel shows the organisation’s usage bar: used, total and percentage. With no balance left:

No tokens available

Purchase a token pack to continue using AI reviews.

Buy tokens goes to the Store — see Plans, credits and extras.

You have to accept before the first use

The first time you write a request — in either feature — the application stops and shows Your acceptance is required. The I have read and accept. Continue button only becomes active once the text has been scrolled to the end:

Scroll to the end to enable the “Accept” button.

I do not accept. Cancel abandons the request and does not use the feature.

The acceptance is recorded with its date, version, language and the hash of the text you read, and you can review it at any time — see Review the terms you accepted.

When the terms change, acceptance is asked for again, on the next use.

What the organisation gets to decide

Two policies under Settings → Policies frame AI for the whole organisation:

AI/MCP documents start as drafts — anything created by an AI agent or through the public API is saved as a Draft for a person to review and send. It is the difference between an assistant that prepares and an assistant that dispatches.

AI access to analytics — whether assistants connected over MCP may read the organisation’s figures. Off by default, with three levels.

Both in Configure organisation policies.

The limits, stated plainly

AI does not sign, and does not send. It prepares and it explains. Publishing still takes a person — and, with the draft policy on, it takes one even when the request came from an assistant.

AI can be wrong. Reading the right document and summarising it badly is a possible failure. For anything with consequences, check the document before you decide.

What you write goes to an external provider. That is what the consent is about, and why it exists.

A signed document can no longer be changed. In the editor: “Not available — the document is already digitally signed and cannot be changed.” In the review panel the conversation stays readable but closed: “This conversation is read-only — the document has already been signed.”

When it is unavailable

Not everything that goes wrong is the balance. The messages tell the cases apart:

Message	What happened
“Too many requests. Try again shortly.”	Rate limit; wait
“Timeout exceeded.”	The provider took too long
“Document is too large for analysis.”	The document does not fit the model’s context
“Content filtered by the provider.”	The provider declined to answer
“Authentication failure with the provider.”	Platform configuration, not yours
“The AI feature is not available at the moment. Please contact the platform administrator.”	No provider is configured

The last two are for whoever administers the platform. The other four depend on the request: rephrase it, split the document, or wait.