Documents Live, a web authoring and publishing system
If you see this, something is wrong
Collapse and expand sections
To get acquainted with the document, the best thing to do is to select the "Collapse all sections" item from the "View" menu. This will leave visible only the titles of the top-level sections.
Clicking on a section title toggles the visibility of the section content. If you have collapsed all of the sections, this will let you discover the document progressively, from the top-level sections to the lower-level ones.
Cross-references and related material
Generally speaking, anything that is blue is clickable.
Clicking on a reference link (like an equation number, for instance) will display the reference as close as possible, without breaking the layout. Clicking on the displayed content or on the reference link hides the
content. This is recursive: if the content includes a reference, clicking on it will have the same effect. These "links" are not necessarily numbers, as it is possible in LaTeX2Web to use full text for a reference.
Clicking on a bibliographical reference (i.e., a number within brackets) will display the reference.
Speech bubbles indicate a footnote. Click on the bubble to reveal the footnote (there is no page in a web document, so footnotes are placed inside the text flow).
Acronyms work the same way as footnotes, except that you have the acronym instead of the speech bubble.
Discussions
By default, discussions are open in a document. Click on the discussion button below to reveal the discussion thread.
However, you must be registered to participate in the discussion.
If a thread has been initialized, you can reply to it. Any modification to any comment, or a reply to it, in the discussion is signified by email to the owner of the document and to the author of the comment.
Table of contents
First published on Saturday, May 9, 2026 and last modified on Wednesday, Sep 30, 2026 by François Chaplais.
You can use AI in Documents Live in two waysyou must have subscribed to Documents Live to use AI.
You ask a simple question from the code Editor
You initiate or continue a chat about a document or a set of documents, either in display mode, or in edit mode if you are an author.
Here is the breakdown.
1.1 Simple question from the editor
1.1.1 What it does
You select some text in the editor and click the AI button. The AI reads your selection, writes a reply, and puts it in your clipboard. A message tells you “The answer is in the clipboard”. You can then paste the reply wherever you like. Nothing is inserted into your document automatically.
Use it for quick, self-contained requests: rephrase a sentence, explain a formula, translate a paragraph, suggest a title, and so on.
1.1.2 What to keep in mind
The AI only sees what you selected. It does not read the rest of your document,
so it cannot answer questions about parts you did not select. If your request needs
context, include it in your selection.
The AI has no memory. Every click is a fresh start. It does not remember your
earlier questions or its earlier answers.
Nothing is saved. The exchange is not stored by the application. If you do not
paste the reply, it is gone.
The reply is written in LaTeX, ready to paste into your document, not as
plain text.
1.1.3 What happens, step by step
You select text in the editor and click AI.
The application sends your selection to the AI service you have chosen in your
AI settings, using your own key. No other service is involved. It adds a set of standing instructions on how
to answer. Those instructions are set by the site’s administrators, not by you.
The AI service writes a reply and sends it back.
The application converts the reply to \LaTeX and copies it to your clipboard.
Your selection ──▶ Application ──▶ AI service
│
Clipboard ◀── (converted to LaTeX) ◀───┘
1.1.4 Who sees what
Sees your selection
Sees the reply
The application
yes (passes it on)
yes (converts it)
The AI service you chose
yes
yes (it wrote it)
Other users of the site
no
no
Your document as a whole
not sent to anyone
—
1.1.5 Privacy compared with ChatGPT or Claude websites
Vendors’ rules change. Check the current terms of the services you use.
When you use a vendor’s own chat website, your conversations are usually kept in
your account there. Here, the application keeps no conversation.
Vendors typically apply different rules to their own websites and to the
connections used by applications like this one, for example about whether your
text may be used to improve their models. The connections used here are generally
treated more strictly, but you should read the terms yourself.
You use your own key, so requests are linked to your account with the
vendor and billed to you.
Only your selection leaves the application, and it goes to a single company, never
your whole document.
If your selection contains confidential information, remember that it is sent to the AI service you chose.
1.1.6 When it does not work
“No API key available…”: add or update your keys in your AI settings. You only need a
key for the service that writes answers.
Nothing happens: check that you actually selected some text.
1.2 Chats about documents
1.2.1 What it does
You open a chat attached to a document, or to one of your document sets, and ask questions in plain language. The AI answers using the content of those documents, and the answer appears progressively as it is written. You can keep asking follow-up questions: the AI remembers what was said earlier in that chat, and you can come back to the chat later.
1.2.2 The idea in one picture
Think of a library with a librarian and an expert:
The librarian (the embedding service) does not answer anything. Their only
job is to find which passages of your documents are relevant to your question.
The expert (the AI service) never visits the shelves. They receive your
question, the recent conversation and the few passages the librarian picked, and
they write the answer.
Both are outside companies that you choose in your AI settings, each with your own key. They can be the same company or two different ones.
1.2.3 Two phases
1.2.3.1 Phase 1 — Getting a document ready (once)
Before a document can be chatted with:
The document must be prepared for AI. If it has not been, you will be told
“This document is not AI ready”.
The document is cut into passages. The cut follows its structure (sections
and subsections), and very small pieces can be grouped together, depending on the
minimum passage size in your settings.
Each passage is sent to the librarian (the embedding service), which turns it
into a list of numbers that captures its meaning. Two passages about the same
topic get similar numbers, even if they use different words.
The passages and their numbers are saved in the application, attached to the
document. This happens once. Later chats reuse it.
Document ──▶ passages ──▶ embedding service ──▶ numbers
│ │
└──────────── saved in the application ◀┘
For a document set, its documents should be prepared beforehand with the set’s “index” action. Documents that are already prepared are skipped.
Only the embedding service sees the full text of your document, and only at this stage. If you change the embedding service or model in your settings, documents need to be prepared again, because numbers from different models cannot be compared.
1.2.3.2 Phase 2 — Each question you ask
You type a question in the chat.
The application checks access: the document must exist, the chat must be
yours, and if the document’s author has restricted AI use on it, only the author
can chat with it.
Your question goes to the librarian (embedding service), which returns its
numbers.
The application finds the 3 best passages. It compares the numbers of your
question with those saved for the passages. The comparison happens inside the
application, and no outside service is called for it. In a document set, all
documents in the set are searched together.
The application builds one message for the expert (the AI service):
the standing instructions on how to answer (set by the site’s administrators);
the 3 passages, as plain text, with their source;
the earlier questions and answers of this chat;
your new question.
The AI service writes the answer, which arrives word by word on your screen.
The application saves your question and the answer in the chat’s history, so
the chat can be reopened later.
┌───────────────────────── saved passages
▼ (in the application)
Question ─▶ embedding ─▶ numbers ─▶ application finds 3 best passages
service │
▼
Answer ◀── AI service ◀── instructions + 3 passages + chat history + question
│
└──▶ shown to you, and saved in the chat history
1.2.4 Who sees what
Embedding service
AI service
Application
Your full document
yes, once, when prepared
no
yes (stored)
Your question
yes
yes
yes
The 3 selected passages
no
yes, in readable form
yes
Earlier questions and answers
no
yes, each time
yes (stored)
The answer
no
yes (it wrote it)
yes (stored)
The numbers for each passage
yes (it computed them)
no
yes (stored)
If the two services are different companies, they never communicate. Each receives its own separate request, and the application is the only link between them. The AI service does not see your whole document, only the 3 passages picked for your question. The embedding service never sees the answers.
1.2.5 Things that help you get better answers
Ask specific questions. Only 3 passages are given to the AI each time. A precise
question finds the right passages more easily than a vague one.
Follow-up questions need to make sense on their own. The librarian only sees
your latest question, not the conversation. “And the second one?” gives the
librarian nothing to search with. “What is the second step of the proof of the main
theorem?” works better. The AI service does see the earlier conversation, but it can
only work with the passages that were found for the new question.
Answers are based on the documents. The instructions tell the AI to answer from
the provided passages. If the relevant passage was not found, the answer may be
incomplete or say it does not know.
Very long chats get shortened. When a conversation grows past what the AI can
handle, the oldest exchanges are removed automatically, including from the saved
history. Start a new chat to change subject.
1.2.6 Privacy compared with ChatGPT or Claude websites
Vendors’ rules change. Check the current terms of the services you use.
Less text leaves the application. On a vendor’s own website you might upload
whole files. Here, the AI service only receives 3 short passages per question, and
only the embedding service sees the whole document (once).
Neither company sees the whole picture. With two different companies, one sees
the document and the questions, the other sees the questions, answers and selected
passages. Neither sees everything the other does.
The history is kept in this application, tied to your chat, not in a vendor
account. It is sent again to the AI service with each question so that it can follow
the conversation.
Vendors’ policies usually differ between their own websites and the connections
used by applications like this one, for example on whether your text may be used to
improve models. Read the current terms of the services you chose.
Your keys link requests to your accounts at the vendors, and usage is billed to
you.
This is not absolute confidentiality: both companies see your questions, and the AI service sees the selected passages. For sensitive material, choose services whose terms you have read, and consider whether the document should be prepared for AI at all.
1.2.7 When it does not work
Message
What to do
“No API key available…”
Add or update your keys in your AI settings. You need one for the AI service and one for the embedding service.
“This document is not AI ready”
Prepare the document for AI first.
“You are not authorized to access this document”
The author has restricted AI use on this document, or the chat belongs to someone else.
“The selected document set is empty” or “not found”
Check the set’s name and content.
“Chat error: …”
The AI or embedding service refused the request (for example, invalid key or exhausted credit). The text after the colon explains why.
2 Create AI accounts and get API Keys
Before using AI in Documents Live, you need to create an account for an AI provider and get an API key for this provider. Documents Live supports currently the following providers:
Anthropic (aka Claude)
OpenAi
Cohere
Deepseek
Google Gemini
Mistral
API keys let Documents Live connect to your AI provider account and use it inside Documents Live.
When an API key has been generated, copy it to a safe place like a password manager, because you will not see it again.
After this is done, go to your account settings at Documents Live. There you will be able to store your API key for usage on the platform.
2.1 Claude
To create an account for Claude, follow this link .
Once this is done, you can get your API key here .
First create your account here . Note: DeepSeek only accepts common domain extensions or Gmail, Outlook or Yahoo emails. .com email adresses work fine.
Once you are connected, you can get an API key from the console.
Below your account identifier in the Mistral AI studio, there is a link that redirects you to your billing settings.
3 Update your Documents Live account
Access you account settings from the top right dropdown menu.
There you can provide your API keys to Documents Live. Theses API keys are stored in encrypted form in the database. What you have to do is select the provider and enter the API key.
Once this is done, you will be back to your account settings. In the "Providers and models" section, you can select
an AI provider
an AI model for this provider
an Embedding provider
an Embedding model for this provider
Below this section in your account settings, is the list of your providers, with test buttons to test the connection for each provider. Below is a screen capture.
Figure 1. The AI section in your Documents Live account settings
I use Voyage for embedding and Claude’s Haiku AI model. Haiku is relatively inexpensive, and it does a very good job at giving well formatted answers.
4 Prepare your document
A document can be prepared for AI (i.e. prepare the RAG text, split and embed it) either from the documents dashboard or within the Inline Editor. We shall detail the first option.
In your documents list, documents that have been prepared for AI have a green badge that says "AI".
Before preparing a document for AI, make sure that the document has been imported recently so that its Documents Live internal model is up to date.
Then, from the "Action" dropdown menu, select the "Prepare for Al" item. That’s it!
5 Chat in the Inline Edit page
In the left sidebar of the Inline Editor is the Chat interface. Open the "AI Chat" tab if needed.
To create a chat, click of the "New chat" button and give it a name.
When you have at least one chat, a select menu lets you chose which chat history you want to use. Select one and click the "Open" button. The chat history is revealed, with a zone at the bottom where you can enter a new question.
Be patient while your answer is being prepared.
After a while you will get your answer.
Note that in the chat history, clicking on a question collapses/expanses the answer to manage space efficiently.
A chat answer has two parts
a rendered part that displays the answer
below, a copy button that lets you copy the LaTeX code of the answer, which you can use later in Documents Live
another button that copies the answer in Markdown format
Unless your document has a very old internal model, the links in the chat answers should lead you directly to the references that have been selected by the AI. Because the links are absolute, they will work in any Documents Live content where you have used the answer LaTeX code. Below is the screen capture of a chat example.
Figure 2. A chat example. Click on the icon above to zoom the screenshot to full window width.
6 Ask AI inside the code editor
You can use AI right inside the code editor.
To do so, type your question in the editor, select it and click on the AI button. The AI answer will be put in the clipboard as LaTeX code, which you can re-use anywhere.