Skip to content دوام Help Center

AI & Automation

The Duwam AI assistant

In one line: a chat panel where you ask about your work in your own words — your tasks and your day, who worked today, how many hours a colleague put in, a customer's number — and the assistant reads live data through tools bound by your permissions, showing anything it would write as a card that does nothing until you press it.

Plan: Available on every paid plan — Basic, Professional, Advanced, Business and the Custom plan — and during the trial; not once a trial ends without subscribing — where the assistant button never appears at all. Using it needs no named permission: anyone who signs in can open it, but every tool inside it is bound by its user's permissions. It also needs switching on by the platform administrator, not by the account owner. Start with Plans & billing.

Overview

The Duwam assistant is a chat panel you open from any page and ask about your work in plain language. It doesn't just answer from its own knowledge — it calls tools that read your live data at the moment you ask, then answers with what the tool returned. Its reach has grown: no longer just tasks and chat, but attendance and hours, pending leave, team workload, an employee's job details, company size, and customer contacts and their numbers.

It is bound by your permissions, literally: it cannot read what you cannot read. The tools themselves check the permission before returning anything; some refuse outright when you don't hold it, and some narrow the question down to you alone and say so in the answer. All of them are sealed inside your company.

Note: the assistant replies in Modern Standard Arabic or Saudi/Gulf dialect, and never uses emoji. Its default answer runs one to three sentences, and it expands only if you ask for detail.

How do I open the assistant and start a conversation?

  1. Press Ask AI in the top bar, at its far end with the other top-bar buttons — the bell and your picture. It shows in every workspace — HR, Customers and Accounting — because it isn't a sidebar row but a button in the bar itself. On narrow screens it drops its text and stays a square with just its icon.
  2. The panel slides in from the edge of the screen, titled Duwam Assistant with "AI assistant that remembers your preferences" beneath it. Type your question in the field at the bottom and press Enter to send (Shift + Enter for a new line).
  3. A new conversation opens on "How can I help today?" with three ready suggestions: "What is on my schedule today?" · "What are my open tasks?" · "What are my high-priority tasks?" Press any of them to send it straight away.
  4. Every conversation saves itself. The panel has three tabs: Duwam Assistant for the running conversation, Threads for earlier ones, and Memory. The + icon in the panel header (New chat) starts a clean thread.

Tip: your conversation stays open as you move between pages, so you don't lose context going from Tasks to the Planner. And if voice dictation is enabled for your company, a microphone button appears in the composer to turn your speech into text.

Closing or deleting a conversation

  • Press the × icon in the panel header (Close), or the Esc key, or click outside the panel — all of them hide it without losing the conversation.
  • To delete a conversation, open the Threads tab, press the Delete icon beside it, and confirm at "Delete this thread?" Deletion is permanent.

What can I ask it?

These are read tools; they change nothing. Ask in your own words — the assistant picks the tool:

About your day and your tasks

  • "What is on my schedule today?" — the blocks pinned to your day, and the tasks due today or overdue. It shows twenty at a time and tells you the full count, so you never read "20" when it means twenty-five.
  • "What are my open tasks?" — your unfinished tasks ordered by due date, narrowable to a single priority.
  • "Any overdue tasks in project X?" or "What is Ahmed working on?" — a search across tasks by name, status, assignee or project. Arabic search folds away hamza, taa marbuta and vowel marks, so «صيانه» finds «صيانة».
  • "What are our projects?" — the company's projects, each with its count of open tasks.
  • "Plan my day" — a suggested daily schedule as a preview only, built from your shift and your tasks, changing nothing until you agree.

About colleagues and conversations

  • "Who is Ahmed?" — find a colleague by name, email or id. If more than one person matches the name, it asks which you mean rather than assuming the first.
  • "Who messaged me?" — your most recent conversations, with a preview of the last message.
  • "What did Ahmed send me?" — the latest messages from a conversation you are a member of (ten by default, thirty at most).
  • "When did X start?" or "What are their hours?" or "Which days do they work?" — the job card: title, start date, working days, daily hours, and teams. Directory data only — no salary, no private details. Your own card is always open to you; a colleague's card is for the company administrator alone, and anyone else is told they don't hold the permission.

About attendance and hours

  • "Who is in today?" or "Who is out?" — today's attendance, sorted: who is on shift now, who has finished, who is on leave, whose rest day it is, and who is absent. Absence here means real absence: someone on approved leave or on their rest day is not counted absent.
  • "How many people worked at least one day last week?" or "Who never worked at all?" — company-wide attendance over a period: how many attended, who never did, and each person's days and hours. It accepts a minimum in minutes ("how many worked more than three minutes").
  • "How many hours did X work this month?" — one person's attendance over a past period: days attended, hours, and a day-by-day breakdown. Hours include overtime and state it separately.
  • For both period questions above: name no period and it reads the last seven days; the longest it will read is 365 days, and anything beyond is trimmed to that.

About leave and team workload

  • "Are there pending leave requests?" — requests not yet decided. Read-only: the decision stays with a human, on the request card inside the conversation. See Leave.
  • "Who is overloaded? Who is free?" — the count of open tasks per person. It counts what actually belongs to them: what is assigned to them, what they are reviewing, and stages they picked up and haven't finished.
  • "How many employees do we have?" or "How many contacts?" — company size: active and inactive employees and teams, and with the permissions, projects, open tasks and contacts. Anything you aren't entitled to see is dropped from the answer and named as dropped.

About your customers

  • "Give me Saad Al-Harbi's number" or "numbers for company X" — a search across customer contacts by name, company name, mobile or email, returning each contact's numbers, email and the company they belong to. Give the name alone, not the whole sentence. It shows ten results by default (twenty-five at most) and states the full count. See Contacts & companies.

Caution: the assistant does not infer or estimate. When no tool gives it a number, it says it doesn't have one rather than deriving it from another — and an account status of "active" means the account isn't disabled, never that its owner has ever signed in.

How does the assistant change anything?

Here two doors part, and the difference between them is whether something passes through a card:

What passes through a card you press

Three things are never written on the assistant's decision alone. It prepares them and shows a card beneath its reply, and nothing happens until you press:

  • A new task — say "get Omar to wash the car within two days" and a card appears headed "New task — not created yet" carrying Task, For, Project, Due and Priority, with Execute and Cancel buttons. A request naming two people becomes two cards, one per task.
  • A WhatsApp message to a customer — a card headed "WhatsApp message — not sent yet" carrying To, Number and Message in full, with Execute and Cancel. If two names look alike it stops and asks which you mean rather than addressing one customer with another's words; and if the contact has no number it refuses instead of offering a button that cannot succeed.
  • A chat message to a colleague — a draft card with Send, Edit and Cancel. Press Edit to correct the text before sending, then confirm.

After Execute the card's buttons switch off and it shows "Task created" or "WhatsApp sent", then settles into a receipt with no buttons that keeps what was done for you to read later. If the send stumbles you read the reason on the card itself: "WhatsApp is not enabled for this company." or "That number is not on WhatsApp." or "Your plan is out of WhatsApp messages for this month." — and the button stays live so you can retry once the cause is gone. "The number is offline — the message is saved and will send when it reconnects." is not a stumble: the message is stored and arrives once when the connection returns, which is why that card closes.

Caution: a card's turn into a receipt is stored in the browser session alone. Open the same conversation on another device, or after closing the browser, and the card is pressable again — and a second press writes a second time: a duplicate task, or a message the customer receives twice. Read the card once, and press once.

What is written directly

These run the moment you explicitly ask for them, with no card, and the assistant states what it did in its reply so you can check:

  • Pinning a task to a time — "block 9 to 11 for task X", which then appears on the Planner.
  • Changing status — the accepted statuses are those of the task's own project, so a project that named its final stage something else accepts it by that name. Marking a task done fires automation and reward rules exactly as if you had changed it by hand — and a change to the same status counts as no event and never pays a reward twice.
  • Changing priority — low, medium, high, urgent.
  • Applying the day's plan — pinning a set of blocks at once after you approve the preview, telling you how many blocks landed and how many failed.

Caution: these tools respect your permissions on the task. An employee edits only what is assigned to them, what they review, or a stage they picked up; an administrator and a project lead edit any task in the company. Creating a task for someone else is limited to the administrator and the project lead — an employee creates only for themselves.

How do I plan my day?

Say "plan my day" and the assistant generates a suggested schedule from your shift, your tasks and the blocks already pinned, shows it as a preview, then asks whether to apply it. On your agreement it pins the blocks to the Planner in one go.

What does it remember about me?

The assistant remembers durable facts about you across sessions, so you don't restate your preferences every time.

  1. After each exchange it automatically extracts what is worth remembering — at most three facts per turn — and saves it.
  2. You can also ask outright: "remember that I prefer to focus in the mornings", and it saves that immediately.
  3. Review everything it has saved from the Memory tab, which carries the line: "These are things the assistant has learned across your conversations. You can forget any of them." Delete any with Forget, then confirm at "Forget this memory?"

Note: the memory is yours alone. The assistant carries 50 memories at a time — most important, then most recent — and those are the ones the tab shows, each up to 500 characters. It does not save transient state ("I'm busy today"), pleasantries, anything sensitive such as passwords or ID numbers, or anything that was its own guess.

Who sees what?

This is the part measured in permissions, not in trust. The assistant's tools are the same doors that guard the screens, not a second copy of them:

The question What it needs
My tasks, my schedule, company projects, finding a colleague by name Nothing
Another employee's job card (title, start date, working days) Company administrator (your own card needs nothing)
All company tasks in a search Company administrator or project lead (otherwise your own tasks)
A conversation's messages Actually being a member of it
Today's attendance for the whole company Manage attendance or View reports (otherwise your own record, and it says so)
Company attendance over a period, and who never worked Manage attendance or View reports — with no fallback to yourself here: the question is about the company, so it is refused
Another person's attendance over a period Manage attendance or View reports (your own attendance needs nothing)
The company's pending leave requests Approve leave (otherwise your own requests, and it says so)
Team workload, and projects and tasks in company size View reports
Customer contacts and their numbers, and a WhatsApp message to a customer CRM access — and the feature enabled for the company

Above all that: a tool belonging to a feature switched off for your company simply doesn't work. If chat is off it reads no messages; if attendance is off it reads no hours — exactly as the screen itself would be closed.

Granting a permission: Employees → open the employee to edit → Permissions → the right group: Attendance for Manage attendance and Approve leave, Reports & billing for View reports, and Customers for CRM access. See Teams & permissions.

What are the usage limits?

The assistant spends on a paid AI service, so four controls work together:

1) A monthly quota for the whole company — the number of messages the company spends per month:

Plan Messages per month
After a trial ends unsubscribed No assistant at all
Trial 50
Basic 100
Professional 1,000
Advanced 2,500
Business 5,000
Custom 20,000

The platform administrator may raise the ceiling for a particular company or open it entirely, so your company's number may differ from its plan's — and the settings card below says which one governs.

A message counts only on a successful reply, and the counter resets at the start of each month. When it runs out you read: "Your company has used this month's Duwam Assistant quota — your admin can add credit or upgrade the plan." The AI buttons inside the customer-reply composer and conversation suggestions spend from the same pot — one balance, not three. See The inbox.

2) Rate ceilings per user — protection against one burst burning the balance in minutes:

Plan Per minute Per hour Per day Company per day
Trial 5 30 50 —
Basic 5 30 50 —
Professional 10 120 600 —
Advanced 15 200 1,000 —
Business 20 300 2,000 5,000
Custom 20 300 2,000 5,000

Crossing them reads: "You have exceeded the Duwam Assistant request limit." and clears once the pace settles.

3) Limits on the message itself — a single message runs to 4,000 characters, and the assistant carries the last 30 messages of the same conversation for context, on top of the facts saved in Memory.

4) An allowance for each person — the company's balance is shared by its members, and each has a daily and a monthly allowance so that one person cannot spend everyone's balance. The default is a share of the company balance, and the company admin holds the largest one:

Role Per month Per day
Company admin The whole company balance A quarter of the company balance
Project manager 30% of the company balance 6% of the company balance
Employee 10% of the company balance 2% of the company balance

Nobody gets less than 10 messages a month and 5 a day. For example, on Professional (1,000 messages) an employee can send 100 a month and 20 a day, and the admin the whole balance and 250 a day. The company admin changes these numbers for each role, or for one person, on the AI usage page. An allowance limits a person and never adds anything above the company balance: the company balance caps everyone together.

The day resets at midnight in the company's time zone, and the month at the start of each month together with the company balance. Someone who reaches their allowance reads: "You've reached today's AI assistant allowance — it resets at midnight." or "You've reached your monthly AI assistant allowance — ask your company admin for more."

Where do I read my usage? Settings → My account → AI usage. At the top, one sentence says whether your pace runs out before the reset, for example: "Heads up: at this pace you'll run out of your monthly allowance on Monday September 28 in the evening, before it resets on October 1." Beneath it are three bars — Today, This month and Company pool — each with when it resets and how much is used. The Ask for more button sends your request to the company admin, once a day.

On the same page the company admin also sees Team usage this month — who used the assistant and how much each person has left, with a button to edit a person's allowance — and Allowances by role. The AI assistant card in the My company group stays for switching the assistant on, with a Usage details link.

How do I add credit? On the same page the company admin presses Get more usage. A window opens to choose a number of messages (100 to 100,000, in steps of 100), and the price shows as you type: $3 per 100 messages — 1,000 messages for $30. Choose how you will pay — Direct payment (transfer or card) or Cash — add a note if you like, and press Send request to Duwam. The request reaches the Duwam team; once the payment is confirmed the messages are added to your company's balance and you are notified. Extra credit is used after your plan's monthly quota runs out and lasts until used — it does not expire at the end of the month — and the Company pool line shows it as Extra credit left. A company has one request under review at a time and can cancel it before it is decided; past requests and their outcome stay under Credit requests. The company pool grows with the extra credit, and the default role allowances grow with it.

Tip: if the assistant button never appears at all, the feature isn't enabled for your company, or your plan doesn't include it. Enabling it is the platform administrator's to do, not the account owner's. The settings card stays visible either way — it is the operating surface — carrying a not enabled badge and a line for the company administrator: "The service is disabled at platform level. Contact the platform administrator to enable Anthropic keys." Only the platform administrator sees an enable the assistant for this company button in its place.

What the assistant does not do

So you don't go looking for what isn't there:

  • It never sends a message itself — neither chat nor WhatsApp. It prepares the card; the press is yours.
  • It never creates a task itself — the card is shown, and the write happens after Execute, not before. If the assistant says "created" before you press, it is wrong; the card is the authority.
  • It does not decide leave. It reads pending requests and names who they belong to; approving and rejecting happen on the request card inside the conversation, by a human.
  • It does not edit employee data — not title, not working days, not hours, not permissions. And it reads no salary and no ID number.
  • It does not clock anyone in or out and does not correct an attendance record. It only reads attendance.
  • It does not create, edit or delete a contact. It searches and shows numbers, and prepares a message — nothing more.
  • It does not touch Accounting, payroll, work orders or campaigns. It has no tool there at all.
  • It never invents an id. A request naming a project or employee id that doesn't exist is refused, and nothing is built on it.
  • It does not work outside your company. Every tool is sealed inside your tenant, and "not found" and "not yours" are said in the same words so neither can be inferred from the other.
  • It does not read a message you deleted for yourself — what you hid from your screen is hidden from it.
  • It does not remember across users. Its memory is yours alone, and no colleague reads any of it.

FAQ

Q: Does the assistant see my colleagues' tasks or chats? A: No. It is bound by your permissions exactly. An employee sees only their own tasks and chats; an administrator and a project lead see company tasks in a search, but reading messages is limited to conversations they are actually in.

Q: I asked it to message a colleague — did it send? A: No. It drafts the message and shows it as a card, and nothing sends until you press Send. The same goes for a customer message: a card with Execute. See Chat.

Q: I asked it to assign an employee a task and it said it had created it — was it? A: Nothing is created before you press Execute on the card. If the card still reads "New task — not created yet", the task has not been created, whatever the text says.

Q: I pressed Execute, then opened the conversation on another device and found the button live — should I press it? A: No. The receipt is stored in the browser session alone, and a second press creates a second task or sends the message again. Check the task or the conversation first.

Q: The assistant set a task to done — does that affect rewards? A: Yes, exactly as if you had changed it yourself: it fires the automation and reward rules. See Tasks.

Q: I asked "who never worked at all" and it said it doesn't have the permission. A: Because that is a question about the company, not about you, and it does not fall back to your own record. It needs Manage attendance or View reports.

Q: I asked for a customer company's numbers and it found nothing. A: Give the distinctive name alone — "company X" — not the whole sentence, because it searches with what you gave it. If you're searching by number, type the number as it is, even in local form; matching is on the number's tail, so it finds one stored in international format. And if it replies that it can't search, the cause is the CRM access permission or the feature being off for your company.

Q: How much of the conversation does it remember? A: The last 30 messages of that conversation, plus the durable facts in the Memory tab.

Q: Why is it suddenly asking me to upgrade? A: Usually because the company's monthly quota is spent. It renews at the start of the month, or you can upgrade for a higher ceiling. If the message is "You have exceeded the Duwam Assistant request limit," that is pace, not quota: wait a little and try again.

Q: How many employees do we have? Why does its number differ from what I know? A: Because it counts your company's active and inactive accounts and separates them, and counts neither deleted accounts nor support accounts. The number present today is something else entirely, and is never derived from this one.

Quick reference

Action Path
Open the assistant Top bar → Ask AI
Send a message Type and press Enter (Shift + Enter for a new line)
New conversation Panel header → + icon (New chat)
Move between conversations The Threads tab
Delete a conversation Threads tab → Delete icon → confirm
Review or forget a memory Memory tab → Forget
Close the panel The × icon, the Esc key, or a click outside
Plan the day Ask "plan my day", then approve the preview
Create a task for an employee Ask in your own words → "New task — not created yet" card → Execute
Send a customer a WhatsApp Ask in your own words → "WhatsApp message — not sent yet" card → Execute
Send a drafted chat message The draft card → Send (with Edit first)
Read this month's usage Settings → My company → AI assistant → usage this month
Grant a permission the assistant needs Employees → edit an employee → Permissions
  • Tasks — what the assistant creates, and whose status and priority it changes
  • The planner — where the day-plan blocks land
  • Chat — the draft it prepares and you send
  • Attendance — the source of the attendance figures it reads
  • Leave — the pending requests it names and does not decide
  • Contacts & companies — the numbers it searches
  • Teams & permissions — the permissions that bound what it sees
  • Plans & billing — the plans that include it and their monthly quota