When papers come into a matter, the chat reacts like a colleague sitting next to you: it tells you what it has seen, asks for your go-ahead before anything that uses more artificial intelligence and, once you give it, splits, files, identifies and analyses what has arrived, describing each step as it goes. After that the conversation carries on like any other.
This only happens with Automatic steps switched on, in the panel options: Panel options. With them off, the paper goes into the vault and nobody touches it until you ask.
Where the papers have to come in
The chat works through papers that come in:
on the client's page, when you drop them on a matter;
on a case file's page, which sends them to its matter;
in the panel itself, when you drop them with a matter or a case file in front of you.
If that matter is on screen, the chat opens by itself. If the papers came in another way (from Correspondence and filings, for example), it tells you the next time you open that matter's conversation.
The greeting and your go-ahead
First it does what costs nothing: it pulls out the text the documents already carry and counts how many pages are scanned. With that it greets you in one sentence, written by the artificial intelligence: how many documents have arrived and what it would take to work on them.
If nobody has written yet what the matter is about, it asks What is this matter about?. One or two sentences are enough; they are saved as the matter's context, which the artificial intelligence reads to split and file better. If you would rather not answer, click Skip this question.
Then it asks for your go-ahead with a button that says what it will use, with no price: Go ahead · uses your provider's AI · 120 pp., with the scanned pages given separately if there are any. If there were older unidentified papers in the matter, you get two buttons instead of one: Only the new ones, with what that covers, and Also the older unidentified papers.
If JUDLEX recommends a different model for this job, there is also a Use … in this conversation button, with the model's name. It only changes the model for this conversation; the rest of the application keeps the one in Settings.
Nothing that costs money starts until you click.
While it works
With your go-ahead, the chat works step by step and describes each one in a short sentence:
It splits each document into the real papers inside it.
It files in their folder the papers it already recognises. This step does not use artificial intelligence.
It identifies the papers that were not recognised during splitting.
It analyses what has come in: what it is, what it adds to the matter, and what it repeats, completes or contradicts in what was already there. That answer carries its citations, like any other: Ask and read the citations.
Each sentence is written by the artificial intelligence, and JUDLEX checks it before showing it: every figure and every name in quotation marks has to come from the step it describes. If not, the sentence is dropped and the application describes the step in its own words. None of these sentences talks about money.
Above the typing box, a strip shows how many documents it is working on and which one it is on, with the Stop button. If you stop it, what has been done stays done and the chat says how much was left. You can keep asking questions in the meantime.
If the connection to the provider drops, the chat says so and carries on by itself as soon as it is back, with the same document; what has already been read is not paid for twice. It only stops if you run out of credit or the key is not valid. And if you close JUDLEX halfway through, it picks up where it left off the next time you open the conversation.
When it finishes
The chat closes with buttons built from the matter's own figures, not from whatever the artificial intelligence says:
Accept the filings, with how many there are: those that can be justified turn green. The rest stay amber for you to look at.
See the unfiled papers, if any did not find a folder.
Open the folder that received the most new papers.
Link them to this case file, if the papers came in from a case file: it adds them as its exhibits.
If the job took a while and you are in another window, JUDLEX tells you when it finishes: Notifications.
What it costs
Describing each step is a call to the artificial intelligence, and so are splitting, identifying and analysing: all of it is paid for with your firm's key, directly to the provider. What has been spent shows afterwards under Settings → History, each item under its own name, and in Daily spend at the foot of the panel.