# Chats, tasks and apps

## Work

### Chats
One chat at a time with each agent. In a chat you can:
- **share files and images:** paste them or drop them in; agents read PDFs, images, Word, Excel and PowerPoint;
- **mention someone:** type `@` for a customer, an agent or a person. The agent gets that record's facts from the database, never from your wording;
- **talk out loud:** press the microphone. The transcript lands in the chat, and the agent can hand work to its "working side" during the talk;
- **follow the work it hands out:** a task the agent opens from the chat shows in the chat as a card with its status
  (live), who has it, and **Open**. When the task waits for you, the card says **Waiting for you**;
- **allow a website:** say "add propertyfinder.ae to the allowed sites", or let an agent ask when it needs a site.
  A card asks **Allow propertyfinder.ae for the office browser?** with **Allow once**, **Always allow** and
  **Reject** (admins; others see "Waiting for an admin"). **Allow once** (from 2.13.0) is for that agent in that
  conversation only, until its work is done and for a day at most; **Always allow** adds the site to Settings →
  Browser at once. The same request waits in the **Inbox**. Sites on the company's own network are never asked for;
- **see how it worked (from 2.13.0):** under an agent's reply, **Worked for 2m 10s · 12 steps** opens the tools it
  used, each with its time, and a link to the whole run;
- **start quickly:** a new chat with Lumate offers four tiles (give a task, call someone, build an app, what can you
  do?) that put a first sentence in the box;
- **start calls:** see [Calls from a chat](/docs/calls#calls-from-a-chat).

**What a talk is like.**
- **Its face:** each agent has its own face on screen, in its own colour and shape. Its mouth moves with its voice, it
  looks toward your mouse, and it tilts its head as it listens to you. It shows during a talk in the chat, in the
  Playground and in the small window while you move around. It is light: it runs only during a talk, and on a PC
  without 3D graphics (or with reduced motion turned on in Windows) a still face shows instead.
- **Its voice:** the agent sounds as it does in the Playground. It uses the delivery from **Settings → Voice** and its own persona, and greets you by name.
- **What it knows:**
  - what Lumate is;
  - who is on the team and who may use the office browser;
  - its open tasks and its recent documents;
  - whether the office browser and WhatsApp are connected.
- **Requests:** what you ask for goes to the working side, which is the same agent in the chat with its tools, and it starts at once. Ask "how is it going?" and the voice tells you what the work is doing now, for example "opening a web page: bayut.com".
- **Replies and documents:** when the work replies, the voice tells you in a sentence or two. It can read you the full reply and any document it made, part by part.
- **Desk phone:** the same happens on a desk phone call you start from the app (**Desk phone**).
- **Browsing:** only agents allowed the office browser (on the agent's page) open websites. Others say which teammate can.
- **Moving around while you talk:** open any other page and the talk goes on. A small window floats over the page with
  who you are talking to, the last words, **Open** (back to the chat) and **End**. Drag it anywhere by its top, or
  shrink it to a dot. Only one talk at a time: end it before you start talking with another agent.
- **"Show me…":** ask the agent to open a page for you — "show me today's customers", "open Jenny's tasks", "go to
  the calls", "open Settings → WhatsApp". It opens the page at once and says which page it opened. It only shows
  pages; it changes nothing. If Lumate's window is minimized or behind other windows, it says so: bring Lumate to
  the front. (On a desk phone call there is no screen, so it cannot.)

### Tasks
A list, a Kanban board (Backlog, Todo, In progress, In review, Blocked, Done) or a table (from 2.13.0; sorts by when
it last changed). The status bar on top counts each status and filters by it. Create one with `c` or **New task**: a title, a description, an agent, a priority, a parent task.

The task page holds the thread, sub-tasks, attachments and the run log. Agents delegate to each other with tasks, and the delegator is woken when the work comes back.

**A task waiting for you.** When an agent asks you a question or needs your decision on a task, the task waits for you
in **In review**; agents never mark it done while they wait for you. (Every agent does this, from the operating rules all
agents share.) The task page shows **Waiting for you** with the agent's last question right above the message box.
Answer there (**Reply**) and the task goes back to the agent, who picks it up at once. A done task can be answered too
(**Reply and reopen**).

### Routines
Recurring work. Ask an agent ("every weekday at 9, list the customers to call") and it sets up the routine, or create one here. Each routine shows:
- its schedule in words;
- the next and last run;
- its history.

You can **Run now** or **Pause** it.

### Documents
(Called Artifacts before 1.6.0.) Everything the team made or shared, as cards with a preview:
- **reports, tables, spreadsheets and pages** the agents made;
- **files people shared** in tasks, chats and customer cards: photos, PDFs, Word, Excel, PowerPoint, text.

Agents' own reference files are not here; they stay on the agent's Files tab.

**Finding things:**
- **Type tabs:** All, Images, Documents, Spreadsheets, Text, each with its count.
- **Search:** by name, task reference or task title.
- **Group by task:** the cards under the task, chat or customer card they came from.

The filters stay in the address, so a filtered view can be shared.

**Each card:**
- shows who made it and when;
- links to **where it was made**: the task, the chat with the agent, the customer card, or the call. A document an agent makes during a chat or task belongs to it, even when the agent does not say so. A file prepared on a call is marked **Playground** (a rehearsal) or **On a call** (a real call);
- opens on a click. An agent's document opens on its own page, with versions, a preview, a download and the changes between versions. A shared file opens in a new tab.

### Apps (from 2.5.0)
Apps are side-panel tools a company builds by asking: "Lumate, build me a payroll tracker", a stock list, a register
of treatments or contracts. They are made of data definitions the company owns; nothing is installed.
- **Build one.** In the menu's **Apps** section, **Build an app** opens Lumate's chat with the request started. Lumate
  asks what it is for, then shows a card with its collections, fields, a few example rows, who may use it (roles and
  agents), each automation (when it starts and what it does) and each widget's template. **Publish** (owners and
  admins) makes it appear in the menu; **Discard** changes nothing. A change Lumate proposes later comes the same way.
- **Use it.** An app's views are tabs: a table (sort, search, add), a board by a choice field (drag a card), a calendar
  by a date, a form, a dashboard (numbers and charts), and widgets (layouts Lumate wrote, such as a payslip: a template
  of plain HTML and CSS filled with the records you may see; nothing in it runs, loads from outside or sends anything;
  links in it are removed). The card you publish shows a widget's template. A record opens
  in a drawer with its history; **Undo last change** steps back one change at a time; deleted records can be restored
  for 30 days.
- **Pages (from 2.12.0)** are an app's own screens with JavaScript, written by Lumate: a delivery board, a planner, a
  calculator, a chart. Ask for one in Lumate's chat ("add a board of today's deliveries to Purchasing"). Two gates keep
  your data in:
  - **Before it is built**, a card says what the page will read and change in the app, which contact fields it reads,
    which of the app's automations it may start, and any outside site it talks to with exactly what goes there. With nothing outside and nothing sensitive it goes
    ahead by itself; otherwise an owner or admin presses **Allow** or **Decline**. Agents can never allow it.
  - **After it is built**, Lumate checks it for anything that could send data out or reach beyond its app, and
    another agent (the Coach) reviews it and lists the risks as high, medium or low. **Preview** shows it once it is
    reviewed (a preview only reads); **Publish**
    adds it as a tab of the app; a high risk blocks Publish until a new version clears it; **Ask Lumate to fix** sends
    the findings back to the chat; **Run the review again** if the review stopped. Every new version is checked again.
  - The page runs in a sandbox: it cannot see Lumate's cookies or pages, open windows or reach any site an admin did
    not allow, and it reaches only its app's records through Lumate, with each person's own rights (members still see
    sensitive fields masked). Pages are not shown in Lumate Desk.
- **Formulas and totals** are worked out on every change (gross − deductions, a linked employee's salary, the sum of
  a contact's payments). A formula that cannot be worked out shows **#ERR** with the reason.
- **Sensitive fields** (salaries, health, ID numbers) are for owners and admins: members see •••, agents never get
  them, nor formulas that read them. A **sensitive app** is for owners and admins only.
- **Agents** read an app by default (nothing for a sensitive one); the agent that built it may change records. Set each
  agent's access in the app's **Settings**. Agents delete records only when you ask in a message.
- **Callers** hear only fields marked "callers may hear" (a clinic's services and prices, open positions), on any call
  Lumate makes or answers; a call about a contact hears those fields of the records linked to them.
- **Automations** start an agent task, a notice, a field change or an email when a record is made or changed, when a
  date comes (checked hourly; for example 60 days before a contract ends), or on a schedule (at most every 15 minutes,
  as a routine). Chains stop after three steps; each automation runs at most 60 times an hour. Add, change, switch off
  or remove them in the app's **Settings → Automations** ("What should happen", then "When it should run"), or ask
  Lumate in chat. **Runs** under each one lists its last runs: when, whether it ran, failed or stopped (and why), the
  record, and whether a page started it.
- **Settings** (owners and admins): who may see and change it, each agent's access, fields (add, relabel, archive:
  archived fields keep their data), automations, revisions with **Compare** (what changed since then, field by field)
  and **Restore** (a new revision; data untouched),
  **Export as xlsx**, **Archive** (off the menu, records kept) and **Delete…** (for good: type the app's name; from 2.14.0).
  To import a spreadsheet, attach it in Lumate's chat and ask: it does a dry run first.
- **Settings → Apps** (owners and admins, from 2.14.0) lists every app, archived and draft ones too, with its records:
  **Open**, **Bring back** an archived one, or **Delete…** one. Deleting removes its records, pages, automations' routines
  and the fields it added to contact cards; export it first to keep a copy.
- **Contacts** can carry fields apps add; they show on the contact's card under "From the apps".
- **Limits:** 50 apps, 20 collections an app, 100 fields a collection, 50,000 records an app.
- **Team packages** carry the apps' definitions; their records only when you choose, never a sensitive app's.