Steps

  1. StartAI goes by a first name that fits the language you use and your store's country, shown at the top of its window and in its greeting. In English it takes the country's name: Noura in Saudi Arabia, Mariam in the UAE, Meera in India, Olivia in the UK, and so on. Pick Malayalam and it takes a Kerala name, Devika (Urdu: Ayesha, Bangla: Nusrat, French: Camille, Arabic: Noura, unless your country already speaks that language and has its own). StartAI Radio's presenter has a name of their own the same way, such as Faisal, or Arun in Malayalam. Ask "who are you?" and it tells you its name.
  2. Each name matches its voice: StartAI has a woman's name and speaks with a woman's voice, and StartAI Radio's presenter has a man's name and speaks with a man's voice. With the browser's own voice, a voice of the same kind is chosen when the device has one.
  3. Press Ask StartAI, the round chat bot button at the bottom corner of any screen, to open the chat in a window; the size button at its top makes the window large or small again. If the button is in the way, press the small – on its corner to tuck it into a tab on the screen's edge; press the tab to bring it back (your choice is kept). Pick the AI model in the bar under the question box: the cheapest are first with the cost of a typical question, and the one you used last is picked for you.
  4. Type a question and press Enter, or press the microphone to dictate it. The answer streams in with the steps the assistant took and any chart or table it drew.
  5. Ask in your own words, for any period: "top selling products this month", "slowest products last quarter", "which category sold most in 2025", "who bought the most of this product", "unpaid invoices over 30 days", "stock value by warehouse" or "the cash account's ledger last week". Periods are worked out in your store's time zone, and every total is added up on the server over all your records, as on the reports.
  6. To ask about a file, press the paperclip (or paste an image, or drop files on the question box) and attach up to 5 files: images, PDFs, text or CSV, Excel (.xlsx) or Word (.docx), 10 MB each. Ask your question, or just press Send to have the file read. Attached files are kept with the conversation; press a file's name to open it again.
  7. Press Talk to have a spoken conversation: speak, and the answer is read aloud as it arrives. Speak again to interrupt; press End talk to stop. Talk voice, next to the model under the question box, picks the voice: any realtime model you connected or the browser's voice; your pick is kept. Talk language, beside it, is the language Talk listens and speaks in: your country's official language until you pick another (Hungarian in Hungary, German in Germany; English in India and where the voice does not speak the language). Your pick is kept for you and StartAI Radio uses it too.
  8. Ask it to add or change almost anything in your own words: "sell 2 brown bread to Al Noor, paid cash", "make a quotation for 10 chairs at 120 for Gulf Supplies", "record October rent of 5,000 by bank transfer", "add customer Al Noor Trading 0551234567", "set the retail price of brown bread to 6". It reads the form of that record, finds the customer, products and other details in your data, checks the change the way the screen does, and asks you if something is missing. Then it shows a card: each line with its quantity and price and the total with VAT for a new document, or each field as it is now and as it will be for a change. Press Confirm, or Cancel. Besides adding and editing records, it can approve or reject requests and leave, make a customer from a lead, post a stock count, invoice a service job or report an invoice to ZATCA. It runs with your permissions and is recorded in the record's history.
  9. Ask it to email someone: "email invoice INV-0012 to Al Noor", "send the purchase order to Gulf Supplies", "remind Al Noor about the unpaid invoices". It takes the address from the customer or vendor (or asks you), writes a short message and shows a card with To, Subject, Message and the document. Press Send and it goes from your store's own mail account with the document's PDF attached, as the Email button sends it. Set up Settings → Email → Outgoing mail (SMTP) first; the assistant tells you when it is missing. It prepares at most five emails per answer and 50 a day per store.
  10. Open Conversations to go back to an earlier chat, or + to start a new one.

Tips

  • The assistant reads only what you can see in the app and can never delete anything.
  • Until your store connects its own AI provider, StartAI answers with free models, for chat and Talk, up to a set number of answers a day; the line under the question box shows how many are left today. Connect a provider in Settings → AI models for faster, more accurate answers.
  • In a demo store opened with your email address, StartAI answers from the demo's own records on the free models: figures, product sales, totals, lists and charts. Talk speaks with Gemini Live, and a Talk session counts as one question. It cannot change anything there or read files.
  • Images and PDFs need a model that reads images (most Claude, GPT and Gemini models do); the assistant says so when the picked model cannot. Deleting a conversation deletes its files too.
  • When the assistant cannot answer a question properly, it says so and notes the question for the StartERP team, who use these notes to improve it. Your store's administrators see them in Settings → AI models.
  • Settings → AI models shows what this month's questions cost and lets an admin set a monthly limit.
  • MCP server in Settings → AI models makes a key for Claude Desktop, Claude Code, Cursor or any MCP client to read this store's data with the same tools. Switch on Allow changes for a key to let that app also add and change records and send emails (without the PDF, which only the app makes): each change comes back as a preview first and runs only when the app calls confirm_action, which it asks you to approve. Nothing can be deleted. The key is shown once; remove it to stop the client.
  • When the store has connected OpenAI with a realtime model (gpt-realtime) or Google (Gemini Live), Talk is a realtime voice call: answers start almost at once and you can interrupt by speaking. Otherwise Dictation and Talk use your browser's speech recognition (Chrome, Edge and Safari). Allow the microphone when asked.

In the app, press F1 or ? on this screen to open this guide next to your work.

This guide belongs to the AI assistant module. Learn more

Related guides

More guides in Dashboards & reports

All guides

Still need help?

Our team answers in Arabic and English by email and phone.