Using the chat
Using the chat
The AI assistant is a chat inside lazyit. Ask in plain words — “which laptops are unassigned?”, “assign MBP-042 to Ana Ruiz”, “take me to the VPN application” — and it looks things up, proposes changes, and opens pages for you. It works with your permissions: it sees and does only what you could yourself, and every change waits for your approval.
The chat appears once an administrator has turned the assistant on and your role includes Use the AI
assistant (ai:use). If you don’t see it, ask an administrator — see Permissions.
[!WARNING] Don’t paste passwords, API keys or other secrets into the chat. What you write, and what the assistant reads to answer you, is sent to the AI provider your administrator configured — see What leaves your server.
Open and close the chat
- Click the chat bubble in the top bar, or press ⌘J (Mac) / Ctrl+J (Windows, Linux).
- On a wide screen the chat sits beside the page, so you can watch the page update. On a smaller screen it floats over the right side; on a phone it fills the screen.
- Press Esc or × to close it. Closing doesn’t stop an answer in progress; the chat picks up where it was when you open it again.
Make the chat wider
- Select Expand (the arrows next to ×); select it again to go back to the normal width.
- Or drag the chat’s left edge. With the keyboard, Tab to the edge and use ← / → (hold Shift for bigger steps), Home / End for the narrowest and widest, and Enter — or a double-click — for the normal width.
A chat wider than normal floats over the page instead of squeezing it. lazyit remembers the width in this browser.
Ask something
Type in the box and press Enter to send; Shift+Enter starts a new line. Press Stop to end an answer early — what was already written is kept.
While the assistant works you see one short line per step, such as “Done: Search assets”. Repeated lookups share a line with a count (“Done: Search users ×5”). Select Show details on a line to see what it found.
Choose the model
Select the settings button under the message box (it shows the chat’s model) before your first message:
- Model — the models your provider offers, or the administrator’s Default model. For one that isn’t listed, type its id in the search box and select Use “‹id›”.
- Reasoning effort — Low, Medium or High, when the provider supports it. Higher effort thinks longer and uses more tokens.
- Temperature — 0 to 2, only for a self-hosted OpenAI-compatible server.
- Auto-approve basic changes — see Auto-approve.
Once the chat has started, model, effort and temperature are fixed (the button shows a lock); start a new chat to change them. Auto-approve can be switched at any time.
The current page
On a page about one item — an asset, a user, an application, a location, a consumable, a purchase, a supplier — the chat shows a chip like About: Asset on this page, so the assistant knows what “this one” means. Only the page address and which record it is are sent, never what’s on screen. Select × on the chip to leave it out of your next message.
Commands
Type / at the start of the message box to see the commands; keep typing to filter, ↑ / ↓ to move, Enter or Tab to run, Esc to close.
| Command | What it does |
|---|---|
/copy |
Copies the whole conversation as Markdown — messages, answers, steps and change cards. |
/new |
Starts a new chat. |
/help |
Shows the commands and keyboard shortcuts. |
/model |
Opens the model picker; /model <id> sets it directly. Only before the first message. |
/auto on · /auto off |
Turns auto-approve on or off for this chat; /auto alone switches it. |
Commands run in your browser and are never sent to the provider. A message that merely starts with a slash (a file path, say), or a command followed by words it doesn’t understand, is sent as a normal message.
Approving what it proposes
When the assistant wants to create, edit, assign, archive or grant something, it shows a card with exactly what will happen and waits for Approve or Reject. The approved change is made with your account, and the page you have open refreshes by itself. How to read a card, bulk changes, passwords and auto-approve are all in Approving changes.
Cards and step lines are written in your language; a sentence lazyit has no translation for is shown in English.
When the assistant asks you for details
When it needs something it can’t find in lazyit — the site new laptops go to, their serial numbers, a date — the assistant shows a short form headed The assistant asks, instead of guessing.
- Fields marked * are required; Recommended ones help it do a better job; Optional ones are under More details.
- Some forms ask for a list, one row per item: use Add a row and the trash icon, within the number of rows the form asks for.
- Lists of sites, categories, models, manufacturers, suppliers or consumables come from lazyit and show only what you can see.
| Button | What happens |
|---|---|
| Send | Your answer goes to the assistant. A missing or invalid field is highlighted and nothing is sent. |
| Continue without | The assistant carries on without the details — it may do less, or ask in words. |
| Don’t ask | The assistant is told not to ask for these details again in this chat. |
While a form waits, the message box is paused with a Go to the form button, and the chat history marks the chat Needs your answer. The form waits as long as a change card (30 minutes by default; it shows Answer by …); if nobody answers, it expires and the assistant stops — send a new message to continue. Answered forms stay in the conversation, read-only.
[!WARNING] The assistant never asks for passwords, keys or other secrets in a form, and lazyit refuses a form that does. Don’t type secrets into a form’s text fields either.
Filling a purchase from a document
On a purchase that has an invoice, an order or a delivery note in its Documents, the assistant can fill the purchase for you: say “here is this purchase order”, or select Ask AI to fill on the document (see Purchases). That button opens the chat with the message written for you — read it and send it; nothing is sent before. If the chat you had open already has messages, the button starts a new chat for it (the other one stays in your history); an empty chat is reused, and anything you had typed in the box is kept after the message.
- The assistant reads the document. This needs Document extraction on in Settings → AI, and sends the whole file to your AI provider, as reading it from the purchase does. It counts against your daily AI budget.
- It asks what the document leaves blank or ambiguous in one form — which supplier (from your list), the reference or the invoice number, a model per line, where the units go, whether they already arrived.
- It proposes the purchase as one card (or a few, for an existing purchase). Nothing is saved until you approve it.
What a supplier wrote is treated as data, never as instructions. Once the assistant has read a document, every card in that chat shows Based on content written by others, and nothing in that chat is approved automatically any more — start a new chat to get auto-approve back. Purchase changes are never applied automatically anyway; see Purchases.
Links and opening pages
After a change, the chat shows an Open ‹item› button. If you ask it to take you somewhere (“open Ana’s laptop”), it opens the page — unless a form on the page has unsaved changes, in which case it shows the Open button instead so nothing is lost.
Links to other websites open in a new tab and show their full address — check it before you click. Images in answers are never loaded.
Sources from the web
If your administrator turned on web search, the assistant can look things up on the internet when lazyit’s records and knowledge base don’t have the answer; with it off, the assistant asks you for the documentation instead. The pages it used appear under its answer as Sources from the web, each opening in a new tab. Other people wrote them: check before you rely on them.
Once the assistant has searched the web in a chat, nothing in that chat is auto-approved anymore — every change needs your approval. Start a new chat to use auto-approve again.
Your chat history
Select Chat history at the top of the chat for your previous chats, grouped by day. Only you can see them — administrators can’t.
- Select a chat to continue it, or New chat to start over.
- Select the trash icon to delete a chat for good. The changes the assistant made stay in the activity log and the AI action log. A chat that is still answering can’t be deleted — stop it first.
Chats are also deleted automatically after the retention your administrator set; the history shows it.
Read-only chats
A chat becomes read-only — you can read it, but not continue it — when:
- an administrator changed the AI provider, or the default model the chat was using (a chat where you chose the model survives a change of default);
- an administrator turned off web search and the chat had it;
- lazyit was updated and the assistant’s tools or instructions changed;
- the conversation grew too long for the model.
Select Start a new chat to go on.
When something goes wrong
The chat says what happened in plain words and offers the next step — Try again, Start a new chat or Reconnect. Closing the chat or losing the connection never loses an answer: it keeps going on the server. For every message, its cause and its fix, see Troubleshooting.