Chat

Chat is the main canvas in LEO, VERA, and ZIA. The product you select determines which knowledge and tools are actually available behind that chat.

Conversations: new, open, and search

New conversation: from the sidebar or by opening /chat.
Recent conversations: remain visible in the sidebar.
Search: can happen through the chat flow or through an inline search workspace.

Titles are generated automatically once a conversation has content, and they are refreshed periodically. If you rename a chat yourself, your version is kept — automatic titles never overwrite a manual title.

Tip:

give important conversations a clear title;
let the assistant generate a better one if the topic of the chat shifts substantially.

Interrupting and resuming answers

Stop generating or the Escape key interrupts a streaming answer.
Interrupted answers are persisted. If you accidentally close the tab, you can reopen the conversation later and see the answer up to the point where it was interrupted.
In long conversations the assistant runs automatic context compaction: older messages are summarized so the chat can continue without starting over. A meta-event in the chat indicates when this has happened.

Product context in chat

Product	What chat mainly draws from
LEO	project documents, personal/shared knowledge bases, broad legal source set, research jobs
VERA	project documents plus Omgevingswet, IPLO, official publications, and RvA
ZIA	project documents, personal/shared knowledge bases, calculator, Tasks, and Word review; no public legal source families (ONS optional)

Nuance:

LEO lite shares chat scope with LEO, but with a smaller source set.

Product scope for chats

Conversation lists are product-scoped:

leo and leo-lite share the same chat scope: leo;
vera uses scope vera;
zia uses scope zia.

That means the same user can see a different chat list in each product.

Personal chat vs. project chat

Personal chat

Without projectId, chat mainly uses:

your personal files/knowledge base;
shared or public knowledge bases you enabled in Knowledge.

Project chat

With projectId, chat mainly uses:

project documents as the primary grounding layer;
project tasks for # mentions;
shared chat history for collaborators.

Move an existing chat into a project

Through Save to project in the chat actions, you can attach an existing personal conversation to a project.

Important:

it keeps the same chat history;
this is one-way in the current version;
project members can open that chat afterward.

`@` mentions and `#` mentions

Mention documents with `@`

Use @ to explicitly include one or more documents.

How it works:

type @ and start filtering;
choose from the list;
use @"..." for names with spaces.

Mention tasks with `#`

In project chat you can mention tasks with #. Tasks are labeled Tasks in the UI; the legacy URL /kanban still works as an alias.

How it works:

type # and pick the task;
use #"..." for names with spaces.

Edit, regenerate, and branch

Under each message you have actions to take the conversation in a different direction without starting over:

Edit one of your own messages to resend a revised version. The original message and its original answer are preserved as an alternate branch.
Regenerate on an assistant answer requests a new alternative. Both answers stay available.
A branch navigator appears as soon as more than one branch exists. Use it to step through alternatives (1/2, 2/2, etc.) without splitting the chat into separate conversations.

For each assistant message we also keep the provider/model attempt history that produced it. You can inspect this on a message, which is helpful when a particular run raises questions.

Attachments: the real lifecycle

Chat attachments are not loose chat-only blobs; they go through the same document layer as Files and project documents.

Uploading

You can add files via:

drag & drop,
the paperclip button.

Choosing a destination

When you are not already in project chat, the shell asks where the upload should go:

personal files,
an existing project,
a new project.

Processing

New attachments move through statuses such as:

pending,
processing,
chunking,
embedding,
indexing,
completed,
failed.

Send blocking

This matters:

while a newly attached file is not yet completed, the composer blocks sending;
that prevents the assistant from answering before the file is actually searchable.

Removing

You can remove an attachment:

from the chat only, or
from the knowledge base as well.

The second option removes it for everyone in that scope.

Inline workspaces

The shell is prompt-first: extra surfaces can open inline inside chat.

Examples:

conversation search,
prompt library,
knowledge controls,
project overview,
project documents,
Tasks (kanban),
research overview or a single research run,
LiDO ECLI positioning map (LEO only).

The benefit:

you stay inside one conversation instead of bouncing between routes.

Slash commands

The composer supports slash commands such as:

/search
/prompts
/knowledge
/project
/documents
/kanban
/research

Note:

/research is only meaningful in LEO context, because research jobs are available there.

Suggested prompts and prompt library

Suggested prompts

Suggested prompts can appear above the composer.

What you can do:

the X button hides them for the rest of the current session;
the refresh button requests three new suggestions;
refresh has a 10-second cooldown.

Prompt library

Through the prompt library you can:

use standard prompts;
filter by tags;
create your own prompts;
manage organizational or public prompts if your role allows it;
import or export prompt sets.

Action cards (Human in the loop)

Sometimes the assistant needs explicit confirmation before continuing, for example before a larger research action or before it writes to an external system.

In that case, it can show an action card with buttons:

each button sends a predefined reply for you;
you do not need to type short trigger words;
after you click, the chat continues immediately with that choice.

Some action cards are prefill actions: they drop a proposed message or email into the input field so you can adjust it before sending. The assistant does nothing on your behalf until you confirm.

Source control inside chat

Chat supports source verification through:

tool cards that show which tools the assistant used and with which parameters;
typed source badges inline for legislation, case law, and official publications. Common types: BWBR (Wettenbank), ECLI (case law), EURLEX (EU law), KOOP (official publications), RVA (Raad van Arbitrage), BIG (BIG register), BAG (Kadaster BAG), NZa, FMS, AP, ACM, UWV, BD (Belastingdienst), LIT (literature / OpenAlex), ECHR, and CORDIS;
additional references such as “Boek N BW” and ordinary law-name clicks, which open the same drawer;
document citations in the form [[file.ext | "verbatim excerpt"]], opening a file drawer with the literal verbatim_text and — when relevant — table cells with colSpan/rowSpan.

Use this actively:

click a badge or [[...]] block to open the drawer;
verify the quote in the source;
treat every answer as a draft until the sources check out.

If an ECLI cannot be resolved, you get a friendlier error instead of a hard 404. Citations without an explicit quote are still linked to the right source where possible.

LiDO ECLI positioning map (LEO only)

In LEO the assistant can open a positioning map that visualizes case law relative to itself:

default view: memo-first with the legal pyramid underneath;
a brick grid of cases with the direction of references between rulings;
by default the map fetches two levels deep (depth-2).

Use it to quickly see whether a ruling has been followed, nuanced, or set aside by later case law.

Voice chat

You can speak your message instead of typing. Chat transcribes your speech automatically and places the text in the input field.

Recording

Click the microphone button next to the send button.
A recording panel appears with a waveform visualizer and timer.
Speak your message.
Click Done (checkmark) or click the microphone again to stop recording.
The transcribed text appears in the input field — you can edit it before sending.
Press Send (or Enter) as usual.

Auto-stop on silence

If you are silent for more than 5 seconds, the recording stops automatically and the text is transcribed. No need to click stop manually.

Appending multiple recordings

Each recording is appended to existing text in the input field. You can record multiple times to build your message piece by piece.

Protection against accidental input

The voice feature has multiple safety layers:

Microphone on by request only: the microphone only activates when you explicitly click the button.
Noise gate: background noise below -40 dB is ignored.
Minimum recording duration: recordings shorter than 0.8 seconds are discarded.
Text confirmation: transcribed text is never sent automatically — you always press Send yourself.

Requirements

A modern browser (Chrome, Firefox, Edge, or Safari) with microphone access.
The microphone button only appears if your browser supports audio recording.
If you deny microphone access, you can enable it later in your browser settings.

DeepThink

Use DeepThink for:

complex analysis,
long or strict instructions,
output that needs to be more stable and careful.

Leave it off when you want quick iteration.

Documentation lookup questions

When you ask “how do I use …” in chat, the assistant can search the docs and cite the right page.