A complete, plain-language guide to Pearl AI, Pearl AI Chat, and Pearl Workstation Agent — from creating an account to running each product day to day. No previous technical knowledge is assumed; wherever a technical word is used, it's explained the first time it appears.
A hosted AI service you can build with, plus a ready-made chat page and a free trial in your browser — no coding required to just start chatting.
Read the guideA chat agent you add to your own website, app or CRM, trained on your own content, with a full dashboard to manage and monitor it.
Read the guideA small desktop application that shows you how your team's work computers are being used. Windows and Mac, billed per computer.
Read the guideAll three products share one account and one sign-in. You set this up once, and it works for whichever products you decide to use — now or later. You do not need to already be a client of Pearl Organisation to sign up.
Return to the products site and choose Sign In, then enter your email and password. If you've forgotten your password, choose Forgot password — you'll be emailed a fresh one-time code, and after entering it you can set a new password and you'll be signed in automatically, without any extra steps.
Pearl Organisation makes several products. They all use the same account, so you can start with one and add another later. Find what you want to do below and start there.
| You want to… | Use | Start here |
|---|---|---|
| Put a chat assistant on your website that answers your visitors 24/7, with no coding | Pearl AI Chat | Adding it to your site, then your website builder, step by step |
| Build AI into your own app, website, CRM or workflow with code | Pearl AI (the API) | Authentication, then your own assistant on your website |
| Chat with AI for your own work: writing, research, documents, slides, spreadsheets, code, pictures | Pearl AI chat page, or PRSXA™ for everyday use | The chat page or PRSXA |
| See how your team works on their computers, with reports and alerts | Pearl Workstation Agent | Installing the agent |
| List your business so buyers can find you, or find trusted businesses and get free quotes | PEARL MARKETPLACE | PEARL MARKETPLACE |
| Custom software, an app or an AI solution built for you | Pearl Organisation's services | pearlorganisation.com |
Once you're signed in, the page called My Subscriptions (My Pearl Products in Pearl Client Workspace) is your home base. Think of it as the control room for everything you've activated. From there you can:
If you visit this page while signed out, you'll be sent to the sign-in page first and brought straight back here afterwards.
On the sign-in and sign-up pages of PEARL PRODUCTS, Pearl AI and PRSXA you can choose Continue with Google instead of typing a password.
Your Pearl Products account also works on PEARL MARKETPLACE and PRSXA™, with the same email and password. Your sign-in email can't be changed, because it is shared by all of them. From your dashboard you can open your marketplace dashboard in one click, and come back the same way, without signing in again.
If you don't have a marketplace business profile yet, one is set up for you automatically. It stays unpublished until you choose to publish it.
Already a Pearl Organisation client? Your products are managed from your Pearl Client Workspace (PCW) instead, and the site takes you there automatically.
On PRSXA™ the plan sets the daily numbers, and the PRSXA plans page shows them live: pictures a day (Free 3, Plus 25, Pro 100, Max 400); Think deeper when you choose it (Free 2 a day, Plus 10 a day, Pro without a daily limit, Max without a daily limit); Deep Research reports (Free 1 a week, Plus 2 a day, Pro 5 a day, Max 15 a day); and Projects on PRSXA Free (up to 2 with 3 files each). Max answers real work on pearl-6.1. Settings > Usage in PRSXA shows what is left today.
Two-step verification adds a one-time code to your sign-in, so your password alone isn't enough to get into your account. It's optional, and you can turn it on or off at any time from the Security tab of your dashboard.
Good to know.
Every product is bought and set up in one of two places. Which one you use depends on how your account was created:
products.pearlorganisation.com) is for everyone. Your products are on My Subscriptions (My Products in the menu), you buy on Choose your product (Account → Subscriptions in the menu), and payments and invoices are on Billing.clientworkspace.pearlorganisation.in) is for existing Pearl Organisation clients. Your products are on My Pearl Products, and payments and invoices are under Billing & invoices.If your account belongs to Pearl Client Workspace and you sign in on the PEARL PRODUCTS site, you are taken to Pearl Client Workspace (or shown a Go to Pearl Client Workspace button). Both places run the same products with the same settings, so everything in this guide applies to both. Where a screen is different, this guide says so.
Every plan purchase also adds bonus AI credits to that product (a share of the price, shown on the checkout page). See Auto-pay, statuses and invoices for top-ups.
| You see | What it means |
|---|---|
| ACTIVE (green) | The plan is running. Any other word (for example CANCELLED or EXPIRED) is shown in grey. |
| Plan | Monthly or Yearly. |
| Assistants / Agents / Workstations "2 / 3" | Pearl Client Workspace: how many you use, out of how many your plan allows right now. The PEARL PRODUCTS site shows Computers n / N for Pearl Workstation Agent. |
| Renews | The end date of the current period. |
| Renews in N days and NN% used | Pearl Client Workspace: how much of the current period has passed (not how much credit you have used). The bar turns amber when 7 days or fewer are left. On the PEARL PRODUCTS site, see the same on Billing. |
| Renews today / Renewal overdue | Pearl Client Workspace: the period ends today, or it has ended and the renewal isn't paid yet. See Renewals & grace period. |
| Activation Key + Copy | Shown on My Subscriptions on the PEARL PRODUCTS site. In Pearl Client Workspace the key is not on My Pearl Products: it is on each product's own configure page. |
| Product | PEARL PRODUCTS site (My Subscriptions) | Pearl Client Workspace (My Pearl Products) |
|---|---|---|
| Pearl AI | Configure Pearl AI (and Buy credits) | Configure Pearl AI |
| Pearl AI Chat | Configure assistant (and the embed code and Buy credits) | Configure Pearl AI Chat |
| Pearl Workstation Agent | Manage workstations (and Add time / computers) | Configure Pearl Workstation Agent |
| You see | What it means | What to do |
|---|---|---|
| Payment couldn't be completed | The payment step failed. No charge was made if the payment step failed. | Press Try again, or use the other method (card or Razorpay). |
| Payment cancelled. | The payment window was closed before paying. | Start again. |
| Please add your billing address below the card details, then press Pay again. | Cards outside India need a billing address. | Fill in country, street, city and postal code under the card, then pay again. |
| Your bank did not approve automatic renewals on this card. Confirming a one-time payment instead… | Your bank refused automatic charges on this card. | Nothing; the payment still completes once. Renew by hand later, or save a different card in Billing. |
| Payment went through but confirmation failed. Check My Subscriptions, or contact support. | Money left your account but the confirmation was interrupted (often a network drop). | Don't pay again. Open My Subscriptions and Billing first. If the product isn't active within a few minutes, email support with the time and amount. |
| Network error — please try again. | Your connection dropped. | Check your internet and try again. |
The Pearl AI, Pearl AI Chat and Pearl Workstation Agent dashboards share the same helpers, on phones, tablets and computers. Knowing them makes every setup quicker.
If you ask Pearl Organisation for help, a member of our team (we call them PEARLians) can open your product's setup page in a limited support view. You never need to share your password.
Pearl AI is an artificial-intelligence service you can talk to, or build into your own software. In plain terms: you send it a question or a piece of text, and it writes back a natural-language answer — similar to the well-known consumer AI chat tools, but made available to you as a building block you can use in your own product, website, or internal tool.
Two ways to use it, depending on how technical you want to get:
The chat page is Pearl AI ready to use, with nothing to set up. Open it with Open Pearl AI on your Pearl AI dashboard (Configure Pearl AI on My Subscriptions or My Pearl Products). Alongside your past conversations and a light/dark toggle, you can:
Everything on the chat page uses your Pearl AI credits, the same as the API.
The Pearl AI chat page at products.pearlorganisation.com/pearl-ai is more than a chat box. Here is what you can do, and where to find each feature.
Click Modes beside the message box (or press Alt+M) and choose how Pearl AI should think. The message box and suggestions change colour so you always know which mode is on, and each chat remembers its mode.
Every mode works at expert level: Coding hands over complete files, exact commands and tests; Research sends several research workers out at the same time (up to 5 with a subscription), one angle each, and writes one answer from all of them; Data & Analysis shows every calculation, with charts and scenarios; Marketing delivers ready-to-publish posts and a calendar; Writing gives the finished piece plus alternatives. The buttons under an answer are next steps Pearl AI can do in one tap.
Answers appear word by word as Pearl AI writes them. While a reply is coming in, the send button turns into a stop button: press it to stop the reply. The part already written is kept, and you are only charged for that part.
Ask Pearl AI to create or redesign an image, or switch to Image mode. It picks the right shape (square for logos, tall for posters, wide for banners) and gives logos a transparent background. While the image is being made, an animated preview shows progress. Under each image you can open it full size, download it, or press Try another version.
Press + and choose Upload a document, or drag a file onto the chat. Pearl AI reads PDF, Word (.docx), Excel (.xlsx), CSV and plain text files up to 20 MB, then answers questions about them: summaries, key dates, risks in a contract, totals in a spreadsheet. Follow-up questions in the same chat keep using the document. Scanned PDFs (photos of pages) are read page by page, up to 40 pages, and each page uses credits.
With a Pearl AI plan you can add up to 40 documents a day, up to 20 MB each. Without a plan (for example during the free trial) it is up to 3 documents a day, up to 5 MB each; a larger file is stopped before it uploads, with a message next to it.
Ask for a chart, for example "show my monthly sales as a line chart", or upload a spreadsheet and ask for one. Pearl AI draws bar, line, area, pie and doughnut charts from your figures. Hover over the chart to see exact values, use Show data to see the numbers as a table, or Download PNG to save the picture. Charts are live: switch between bars, lines and pie, see the table, and download the figures as CSV. Ask about a loan, an EMI, a SIP or a margin and you get a calculator with sliders whose results update as you change the numbers.
Press Listen under any reply to hear it, or turn on Hands-free to talk with Pearl AI out loud. In Voice settings you can switch on Natural voice (HD) and choose from ten human-sounding voices. The HD voice is part of a Pearl AI plan with AI credits and uses the credits for each minute of speech (a Pearl AI plan that counts monthly usage instead does not include it); if it is not available, the standard voice of your browser is used instead.
Point at one of your earlier messages and press Edit. Change the text and press Save & send: that message and everything after it is replaced with a new answer. The earlier answer is kept: arrows under the answer (‹ 1 / 2 ›) switch between the versions, and Use this version brings one back. The same happens with Try again.
Under each reply, PDF and Word save that answer together with your question. To save the whole chat, press Export in the top bar. Images and charts are included. Word files open in Microsoft Word.
Press Share in the top bar and then Create public link. Anyone with the link can read a copy of the chat; photos and documents you uploaded are never included. Use Update with latest messages to refresh the copy, or Stop sharing to take the page down. Shared pages are hidden from search engines.
Activity & logs in the sidebar lists your conversations with the number of messages and the credits used in each. Open one to see every message, its cost and response time, or export everything as a CSV file.
Since 6 November 2026 every kind of use has its own name: your chats, the API, your assistants (shown as Try it or by the API key they used), Read aloud clips, audio overviews, live calls with their minutes, the WhatsApp business assistant and the AI used inside your artifacts. By source compares your messages from chat, the API and assistants, and the CSV download uses the same names.
Press Ctrl+K (or Alt+K) to search your chats and every action, or type / at the start of a message for more than 100 ready commands. Alt+, opens Settings, with about 30 sections from Profile, Security and Devices to Usage, Billing, Capabilities, Reflect, Time and focus, Appearance, Accessibility, Notifications, Data & privacy, Shared links and Connected apps (more about Settings). Other shortcuts: Alt+N new chat, Alt+M modes, Alt+O outline of the chat, Alt+A Apps, Alt+G Agent mode, Alt+C Code Studio, Alt+L Library, Alt+P PDF Studio, and Alt+/ for the full list.
Press Alt+A for 26 ready starting points, including Automations, Deep Research and To-dos in the Automate group. Fill in a short form and Pearl AI does the rest: a website, landing page, app or tool, game, dashboard, slide deck or React component; Design Studio for social posts, posters, flyers, banners, business cards and more; a logo; a document or report, article, email or business plan; Data Studio, reading a document, research or comparing options; running Python or JavaScript, fixing code or drawing a diagram; making an image or editing a photo.
Websites, apps, designs, diagrams, React and React Native code open as live previews beside the chat. Preview on real phone and tablet sizes (iPhone, Pixel, Galaxy, iPad) or type your own size. Brochures, company profiles, reports and proposals come out as real A4 pages with a cover and page numbers, ready to download as a PDF. Websites and designs use real photos from Pearl Photos instead of empty boxes, and Pearl AI is built not to invent figures, phone numbers or emails.
Paste a public link and Pearl AI reads that page, and up to two more pages of the same site such as its About page, then answers from what the site actually says. Pages that need a sign-in cannot be read.
The Answer options button in the message box sets how Pearl AI answers. Thinking: Auto thinks deeper by itself when a question needs it, Think deeper always reasons step by step first (best for maths, logic, code and analysis) and shows a summary of its reasoning as it works, and Answer fast skips the extra thinking. Style: Normal, Concise, Explanatory, Formal, or My style, learned from a sample of your own writing. Formulas appear as real maths, and answers that use the web cite each source inline.
On paid plans, press the wand button in the message box (or Alt+I; Ctrl+Option+I on a Mac) and Pearl AI rewrites what you typed into a clearer, more complete request. You see it first: Use this, Edit it, or Keep mine. Details only you know appear as [placeholders] to fill in, and it never answers or sends anything by itself. Under any reply, Try again also offers Think deeper and Search the web, along with shorter, more detailed, simpler and more formal versions.
Spreadsheets in the chat can be edited: click a cell and type, or double-click it (on a phone, tap the selected cell again), and start with = for a formula. Enter saves and moves down, Tab moves right, Delete clears, Ctrl+Z undoes and Row adds a row. Totals recalculate at once and Download Excel includes your changes. Reset puts the sheet back the way Pearl AI made it. Slide decks have a theme picker next to Present; the theme you pick is used for the PowerPoint and PDF downloads too, and the first swatch puts back the look the deck was made with. Your changes are kept in this browser only, never for temporary chats, and are cleared when you sign out.
Ask Pearl AI for a budget, a tracker, an invoice, a schedule, a loan plan or any Excel file, and the answer becomes a real spreadsheet right in the chat, with tabs for each sheet and formulas that work. Click a cell to see its formula, use the arrow keys to move around, and press Download Excel to get an .xlsx file whose formulas stay live, so changing a number updates every total. CSV and Copy give the values. Ask in the chat to add columns, rows or sheets.
Ask Pearl AI for a presentation, slides or a pitch deck, or open Apps > Slide deck, and the answer becomes real slides right in the chat. Click through them with the thumbnails, press Present to show them full screen (arrow keys to move, N for speaker notes, F for full screen, Esc to stop), and download them as PowerPoint, with text, tables and charts you can still edit, or as a PDF with one slide per page. Ask in the chat to change any slide, the theme or the brand colour.
On paid plans you can keep related chats together in a project. Open Projects in the sidebar, create a project and give it instructions, such as who you are, the goal, the tone and facts to keep in mind, and every chat in the project follows them. Add files such as a brief, a price list, a style guide or data, and Pearl AI reads the parts that matter for each message. A new chat started while a project is open goes into it; you can also add the open chat to a project or move it out. Each project can hold up to 25 files. Deleting a project deletes its files but keeps its chats.
On paid plans Pearl AI remembers lasting things you share, such as your role, your company, your projects and how you like answers, and uses them in new chats so you explain less. With a Pearl AI subscription it keeps up to 3,000 notes, by topic (about you, work, people, projects, preferences, places), and each answer uses the notes that matter for that message; a correction changes the old note. Nothing is deleted because it is old: near the limit, related notes are merged into one, keeping every name, number and date, and Condense now does it at once. Pearl AI also finds your earlier chats by what they were about, even months back. Open Memory in the sidebar to see everything it remembers, add something yourself, forget any item, delete everything or switch Memory off. Passwords, health, money and other private details are never kept unless you ask, and temporary chats are never remembered.
Every page, tool, game, dashboard, website, design or diagram that opens in Pearl Canvas is kept as an artifact with every version, so you can open it again on any device. Under its name in the canvas, Saved opens three tabs: Versions (look at any earlier version and Restore it; restoring adds a new version, so nothing is lost), Abilities and Share (a link anyone can open, through Publish). Type a new name at the top of that menu to rename it. Your artifacts (the Artifacts button at the top of Creations, or type /artifacts) lists everything with search and kinds (Pages, Websites, Apps, Designs, Documents, Diagrams, Code), where you can open, rename or delete an artifact or go back to the chat it came from. Nothing is kept from temporary chats.
Abilities are off until you turn them on for an artifact. Shared data lets a page keep records that everyone who opens it sees, such as a poll, scores or a guest book. Personal details, such as names and emails from a sign-up form, go in a private collection: visitors can add to it but see only what they added, and you download all of it as a spreadsheet file from Abilities. You choose whether only the person who added a record may change it, and you can delete all records at any time. AI inside this page lets the page ask the AI, for example to grade a quiz or write a reply; each answer uses your AI credits, up to the daily number you set for the page. Both work in the preview and on the shared link (not on a link with a password). Visitors never see who you are, and a page never reads your chats, memory or connectors. Ask in the chat to build it, for example “make a guest book page everyone can sign”.
Click the globe button in Pearl Canvas to put the open page or site at a public link anyone can open without signing in. Published pages run sealed off from your account, so they can never see your chats or details. Made changes? Click Update to put the latest version at the same link, or Unpublish to switch the link off. With a Pearl AI subscription your account can have up to 100 published sites and up to 5 own domains (3 sites without one); see Your own domain and Settings > Publishing below. Pages must not ask visitors for passwords or card details or pretend to be another person or brand; pages that break these rules are taken down.
Click Temporary chat under New chat (or type /temp) for a private conversation. It is not kept in your chat list or Library, is never searched later and cannot be shared, and it is deleted as soon as you leave it or, if you close the page first, the next time you open Pearl AI.
Refer to an earlier conversation, for example "the logo ideas from last week" or "continue the business plan we started", and Pearl AI looks through your own earlier Pearl AI chats, answers from what they contain and shows the chats it used as links under the reply. It only ever sees your own chats. Switch it off any time under Answer options, Search past chats.
When real computation makes an answer better, Pearl AI writes Python (with numpy, pandas, matplotlib and scipy) or JavaScript, runs it in a sealed sandbox in your browser, shows the output and charts under the code, and then answers from the real results. CSV, Excel and text files you attach in the chat can also be read by that code. Pearl Canvas keeps every version of a page: click the version number to open, compare or continue from an earlier one. Files the code saves, such as a cleaned CSV or an Excel workbook, appear under it as downloads, and every table in an answer can be copied or downloaded as CSV or Excel. Code running can be switched off in Answer options. When a program asks a question (Python input or a password prompt, JavaScript prompt or confirm), the question appears inside the output with a box to answer, and Stop ends it at once. Next.js and React projects preview too, and an error shows in a clear box instead of a blank page.
Press Alt+G and describe a goal. Pearl AI writes a step-by-step plan and nothing runs until you press Run plan. Steps that change something, such as saving to memory or creating an automation, always wait for your OK.
Run a prompt every day, week or month at the time you choose, with or without live web search, and get the result by email, for example a morning news summary, or a weekly business report from your Google Analytics, Search Console and YouTube when Google is connected. Automations are included with your Pearl AI plan, and each run uses credits like a message.
Paste a very long text (more than 8,000 characters) and it becomes a “Pasted text” card above the message box, so you can still type your question: press View to read it or the cross to remove it. Up to 5 cards go with one message, and one message can hold about 58,000 characters in all. On a computer, press + and choose Take a screenshot, pick a window, tab or screen, and one picture is added to your message; the sharing stops straight away. On a phone, take a normal screenshot and add it with + and Upload an image.
Besides PDF, Word, Excel, CSV and text, Pearl AI reads PowerPoint slides with their speaker notes, OpenDocument files (.odt, .ods, .odp), Rich Text (.rtf), EPUB e-books, saved emails (.eml, with the names of their attachments), log and TSV files, in the chat, in project files and in Knowledge. Long PDF and Word files are read from the first page to the last, up to about 2 million characters, and each answer uses the parts that match your question; answers about a PDF can name the page. If a file is even longer, Pearl AI tells you which pages were read. Pictures inside documents, copy-protected e-books and old .ppt, .doc and .xls files are not read; save those as .pptx, .docx, .xlsx or PDF first.
Paste a YouTube link (up to 3 in a message) and ask for a summary, key points, notes, a quiz or what was said about a topic. Pearl AI reads the title, channel, description and spoken words, and gives times such as [04:35]; follow-up questions work without pasting the link again. It reads the words, not the pictures. If a video has no captions it says so and uses the title and description; private, removed or age-restricted videos cannot be opened.
Web pages, search results, files and your Gmail, Drive, Calendar and Sheets results are treated as information only. If a page or an email contains text that tries to give Pearl AI orders, it is never followed. When you only ask to read something, Pearl AI does not make email, meeting or sheet drafts on its own; ask for them directly. Nothing is sent, booked or written until you press Approve.
Save text you type often as your own slash command: open Settings > My commands and press New command, choose Save as command at the bottom of the / menu, or press Save as a command at the top of any chat to turn its first message into a command. Type / and your commands come first; {} marks where the cursor goes. Up to 50 commands, on every device.
Everything Pearl AI made for you (pictures, documents, slides, sheets, websites and code, designs, charts, quizzes) and the files you shared, from all your chats, in one place: press Creations in the sidebar or in the Library, or type /creations. Pick a kind or search, and click an item to open it at that message in its chat. Temporary chats are never included.
With a Pearl AI plan: move a chat into a project with Move to project in a chat’s three-dot menu or Move in the Library (many at once); edit a memory with the pencil, download all memories as Text or JSON, or Import them from another assistant (you see the list and choose before anything is added; private details are left out); and Automations can run one time at a date you pick, every 1 to 12 hours, or tick Only tell me when something changes to hear only when the result is different, starting with What changed. All automations together run up to 30 times a day.
Under a finished answer, Google Docs saves it as a new Google Doc in your own Drive, and Google Sheets saves its tables as a new sheet (connect Google in Settings > Connectors with Google Drive switched on; up to 20 files an hour). Double-check under the latest answer (or /doublecheck) checks the facts again against current web sources and tells you what it corrected. To-do under any answer saves a to-do with its title filled in and an optional reminder time.
What’s new at the bottom of the sidebar (or /whatsnew) lists the latest features with a Try it button; a red dot means something you have not seen yet. While an answer is being written, the send button turns into Stop. In Pearl Canvas, the full screen button or Alt+F switches at once, and Exit full screen brings it back beside the chat. Your day now sits under the greeting, just above For you.
Select any words in an answer (with the mouse, the keyboard or your finger) and a small bar appears next to them. Ask about this puts the words in the message box as a quote, so you can type your question about exactly that part. Rewrite this part asks “How should it change?”: pick Shorter, Simpler, More formal or Fix it, or type your own words and press Enter. It is sent as a new message and you get the whole answer again with only that part changed; your earlier answer stays as it was. Explain explains just that part in simple words, and Copy copies only the words you selected. If something is already typed in the message box, send or clear it first, so your draft is never lost. With the keyboard, select with Shift and the arrow keys, press Tab to reach the bar and Esc to close it; on a phone the bar shows under your selection.
Open Settings > Websites for search (right after Capabilities) to choose up to 10 websites that web search looks at first, under Search these first, and up to 10 websites it never uses, under Never use these. Type an address such as bbc.com or paste a link and press Add; it is cleaned to the plain address for you, and a message under the box tells you if something is not a website. Pages on sub-sites count too, so wikipedia.org also covers en.wikipedia.org. Your lists are saved with your account and work in your chats, Deep Research and automations; pages from websites you never use are never used or shown as sources. For one question only, say it in your message with the full website address, for example “search only on gov.in” or “don’t use wikipedia.org”. Choosing websites never runs extra searches or uses extra credits.
You do not have to wait for an answer to finish. Type your next message and press Enter: it waits above the message box (“Sends when this answer finishes”) and is sent by itself when the answer is done. Edit puts it back in the box, Send now stops the current answer and sends it straight away, and the cross removes it. Press Enter again to add more to the waiting message. If you stop the answer, or it does not finish, your message waits until you press Send now. A waiting message stays with its own chat and comes back after a page refresh in the same tab. On a phone, tap the small button next to Stop.
Deep Research reports and long answers with several headings have an Open as report button. It opens the answer full screen like a document: the contents on the left take you to any section and show where you are (on a phone, tap Contents at the top), the A buttons change the text size, Copy all copies the whole report ready to paste, and Print / Save as PDF prints only the report; choose Save as PDF in the print window to get a PDF. Deep Research reports also show How this was researched: the goal, the planned searches, the sections and the number of sources. Press Esc or the X to go back to the chat.
Under a long answer, a Deep Research report or a document card, press Audio overview to hear a spoken overview of 2 to 5 minutes. It plays with pause, 10 seconds back and forward and speeds from 0.75x to 2x, shows the written transcript, and can be downloaded as an MP3. Pressing it again on the same answer plays the same overview again without using more. Audio overview is part of a Pearl AI plan with AI credits: 15 a day, paid from the credits (a plan that counts monthly usage instead does not include it), and overviews are deleted 60 days after they are made.
Ask for a practice or mock test (JEE, NEET, SAT, any topic or your own notes), or type /practice. You get a timed, exam-style test with sections, a list of questions showing which are answered, marked or not visited, and Mark for review. At the end you see your marks, accuracy, time per question, a table by section and every answer explained, with Retry wrong ones. Up to 150 questions and up to 300 minutes; your last 40 results stay on your own device.
Ask about the weather or the time somewhere and the answer shows a live card: the weather now, for the next 24 hours and for 7 days, with a C/F switch, or live clocks with the time difference between places and a converter. If live data cannot be fetched, you get a normal answer from a web search instead.
Before a big job where a wrong guess would waste effort, Pearl AI may ask 1 to 3 quick questions in a panel above the message box, each with tap answers and the recommended one first. Tap an answer or type your own, then Continue, and the work starts at once.
Slide decks are planned first, then every slide is written at the same time by separate helpers and checked, and weak slides are made again. Other big answers are reviewed and rewritten when they fall short. A Pearl AI plan has up to 16 helpers at once.
Share your project as a zip or code files and ask for a change. The answer gets a card, Checked against your files, with a ready installer to download, the commands to copy and every change listed. The installer checks first, keeps a backup when it installs, and has an undo step. Nothing runs on your server by itself. A Pearl AI plan includes 30 a day.
Questions about things that change, such as who holds a post, prices, results or news, always get a fresh web search, and up to 5 sources show under the answer. They stay with the answer when you open the chat again.
A PDF or A4 document shows as a card while it is written (“Writing your document”). Open the document shows the pages, with PDF and Word downloads.
Screen readers say “Answer ready” when an answer finishes, the + menu works fully from the keyboard, focus stays inside open windows, and wide code can be scrolled with the arrow keys. Settings > Accessibility has Reduce motion, Higher contrast, Underline links and Larger buttons.
Ask “show me photos of the Taj Mahal”, “pictures of snow leopards” or “Eiffel Tower images” and a photo card under the answer shows up to 6 real photos from Wikipedia and Wikimedia Commons, the main photo first, each with who took it, its licence and a link to its page. “Who is” and “where is” questions show the article’s photo. These photos are never made by the picture maker. Click any picture in an answer to see it large, with previous and next.
Pick, paste or drop up to 4 pictures in one message. Mark up on a picture opens an editor with Draw, Circle, Box and Arrow in red, yellow or blue, with Undo and Clear; the marks and a close-up of the marked spot go with the picture, so Pearl AI knows exactly what you mean.
Press + and choose Scan a document. Use the camera (or choose photos on a phone), pick the Document look (a clean white page) or Colour, turn, move or remove pages, up to 20, name the file, then Download PDF or add the pages to the chat.
Ask for places in an area and get a card with a map and pins (“show more” adds new places). Currency questions show a converter with daily rates; share and index questions show a live card (not investment advice), and “add it to my watchlist” follows up to 15 shares. Ask for a score and get a live card for cricket, football, basketball, American football, F1, tennis, hockey and baseball.
Paste a link to a public GitHub project, folder or file and the answer comes from the real files, with exact lines when you ask about a file. Share your own website address and Pearl AI checks it for real: whether it loads, the security certificate, links, phone setup and speed, and reports only what it found.
The GitHub coding agent works on every account: connect GitHub once in Code Studio (or type /github), pick the repositories it may use, give it a task, see the checks and send the changes as a pull request on a new pearl/ branch for you to review.
Share a chat with chosen people only. In Publish, Who can open it can be anyone with the link, only people with a password (access lasts 30 days, and a new password ends old access) or specific people (up to 25 email addresses, each confirming with a one-time code). A shared chat has Continue this chat, which opens a private copy.
Press Ctrl+Shift+F to search all your chats and everything made in them. In a long chat, Outline lists every question and heading. Under a long answer, Turn into makes it a slide deck, sheet, quiz, web page, infographic, summary or email.
On a computer, press + and choose Share your screen live. A red bar with a preview and Stop shows above the message box, and every message you send includes a fresh picture of your screen until you stop.
Add a .docx or .xlsx file and say what to change. You download the edited file with its formulas and styles kept. Up to 30 a day on any Pearl AI plan (it uses your plan, or AI credits on a credits plan); files are kept for 24 hours.
With a Pearl AI subscription, Apps > Media has 8 templates: poster, social post, YouTube thumbnail, festival greeting, product photo, sale banner, invitation and profile cover. Ask for a carousel or several pictures and get up to 4 at the exact size you asked. A photo you share for a page or design is used as it is, never swapped for a stock photo.
Switch on Study for a chat (or type /study) to learn step by step. Forms in answers show errors under each field and send your answers with one press; checklists keep their ticks.
Share a project with people in your team as Viewer or Editor; they use its instructions and files in their own chats, and chats stay private. Share your saved commands with the team from Settings > My commands. Browse directory in Settings > Connectors lists 29 ready apps to add as custom connectors (with a subscription).
Allow notifications and your computer or phone tells you when a reminder is due, Deep Research is ready or a background task has finished. Voice typing works in the main language of almost every country (73 languages) and in more browsers, up to 60 seconds a recording. Charts, tables and pictures in an uploaded PDF are read too, up to 8 pages.
Every new chat opens with your day at a glance, with nothing to set up. You see your open to-dos for today (overdue ones first, then today’s, then a few without a time) and can tick one off with a single tap. Below them, Pearl AI suggests up to three things from your chats of the last three days that were left unfinished, such as a draft still to be sent or a decision still open. One tap opens that chat with a ready message in the box; it is never sent until you press send. The list is made fresh each morning from your own chats only (temporary chats are never used), nothing is invented, and personal subjects are left out. Only you see it and nothing is emailed. Use Hide for today, or turn it off and on again in Settings > Capabilities > Your day on the start screen.
Under Your day, the first cards in For you are new pieces of the work you do again and again with Pearl AI, such as this week’s posts, the weekly report or a follow-up email, each marked From your chats with a ready message. A tap puts the message in the box; nothing is sent until you press Send. They are written from your own chats of the last few weeks and what Pearl AI remembers about you. Temporary chats, chats marked “Don’t remember this chat” and paused memories are never used, and sensitive subjects are left out. Press the cross on a card to hide that idea for good, or turn them off in Settings > Capabilities > Ideas from your chats in For you. Until you have a few chats, For you shows ideas for your work profile.
If your credits or your plan’s allowance run out, or something is part of a higher plan, the answer area shows one card: what ran out, when it comes back, and buttons for the real next step, such as Buy credits, See Pearl AI plans or See usage. Your message stays in the box, and Retry appears as soon as it can work.
Under each answer, Copy, Listen, Try again and Good or Not good stay in view; every other tool, such as Branch, Double-check, To-do, PDF or Word, is under More. On a computer the top bar keeps Outline and Share, and the rest (voice settings, Hands-free, light or dark, Export, Usage) is under More. Press the up arrow in an empty message box to edit your last message.
Searching the chat list also offers Search messages, files and creations, so one search covers your chats and what is inside them. To move a chat out of the way without deleting it, choose Archive in its three-dot menu: it waits under Archived at the end of the list, still opens, and still shows in Library and search. Choose Unarchive in the same menu to bring it back.
If your internet drops while Pearl AI is writing, the answer keeps growing in place as soon as you are back online, with a short note that it is still being written. The full saved answer replaces it when it is finished. If Pearl AI cannot be reached when the page opens, it tries again by itself and then shows Try again; you stay signed in.
Each chat in the list shows just its name. Point at a chat (on a phone the button is always there) and press the three dots: Pin, Rename, Mark as unread, Move to project, Archive and Delete (Delete asks first). Esc closes the menu and the arrow keys move through it.
Mark as unread puts a dot and a bold name on the chat on all your devices, so you remember to come back to it. Opening the chat marks it read again, and Mark as read in the same menu does it without opening. Marking the chat that is open keeps it unread until you open another chat. If it cannot be saved, a short message shows right under that chat.
With the sidebar closed, pointing at the sidebar button slides it open over the chat; it stays open while a chat’s menu is open and slides away when you move off it.
When your research request leaves out something that would change the report, such as your city, budget, time frame or who the report is for, the plan comes with up to 3 short questions. Answer any of them and press Update the plan (or Enter) to get a plan made with your details, or just press Start research.
While a report is waiting or being researched, press Add a note on it to steer it, for example “focus on India” or “prices in rupees”. You can add up to 5 notes; they show under the report, and the last searches and the report follow them. Once the report is being written, ask in its chat instead. If a note cannot be added, the reason shows right under the note box.
If an answer is going the wrong way, you do not have to wait for it to finish. Type what you want instead, for example “shorter”, “in Hindi” or “only for Dehradun”, and press Change direction next to the message box (or Ctrl+Enter). Pearl AI stops where it is, keeps what it already wrote, and follows your new direction straight away. Pressing Enter while it writes still sends your message after the answer finishes.
A chat with several big answers opens in a moment with your newest question and answer. Earlier messages come in as you scroll up, or press Show earlier messages. Nothing is deleted: export, print and the outline still include the whole chat. Long answers are drawn a little more gently, so phones and tablets stay cooler while the text arrives at the same pace.
Short notes, such as a mode turned on, move clear of an open menu instead of covering it. A message about a button in the message box shows above the box, never over what you typed. Menus always stay on the screen, and only one menu is open at a time. When the message box is short of space, Change direction becomes a round button with a turning arrow. On phones, Pearl AI notifications show its mark in the status bar.
A small site gets a quick page link right away (up to 4 MB, one page up to 2 MB). A bigger site opens the full Publish window by itself, which keeps sites up to the size your plan allows and adds versions, your own domain, forms and stats. Pictures saved inside the pages are what usually makes a site big: ask in the chat to make them smaller if you get near your plan's size.
Node.js projects run in the preview with many common packages, such as express-validator, joi, zod, axios, dayjs and winston. If a project uses a package the preview cannot run, you see a short note naming it instead of an error, and Download zip runs the whole project on your computer. Ask for the full project and every file comes as its own block. Published sites never show developer error cards to your visitors.
If you lock your phone, switch apps or lose the connection while an answer is being written, the answer keeps growing when you come back. If it really could not finish, it says so and offers Retry.
Code Studio fits small screens, and its three-dot menu has Search in all files, Ask AI, GitHub and Publish. Scrolling never gets stuck at the end of a code block. Hint text in typing boxes is lighter, and in Settings > Publishing > Manage your domain shows on one line and Back takes you to the Publishing list.
Every slide deck has Edit slides: change the text of every layout (title, points, columns, numbers, quote, chart labels and values, timeline, cards, table), the notes in What to say on this slide, the order (drag on a computer, or Earlier and Later), Duplicate and Delete, with Undo and Redo. Theme offers As made, Ocean, Midnight, Forest, Sunset, Royal and Mono, and downloads use it too. Download PowerPoint, PDF, one slide or all slides as pictures, or copy the outline. Up to 40 slides. Your edits are saved in this browser and the card shows Edited by you · Reset. While presenting, press N for Presenter view (notes, next slide and a timer), B for a black screen and ? for the keys.
Above every spreadsheet: Find in sheet, Formulas (show the formulas), Chart (bar, line or pie from the cells you select) and Full screen. Click or tap a column letter to sort, filter rows by that column, select it, chart it or fit its width; sorting and filtering only change what you see. Select cells to see Sum, Average, Count, Min and Max under the grid. Copy, paste and clear ranges, with Undo and Redo. Your edits are kept in this browser and go into Download Excel and CSV; Share live lets your team edit together.
Press Open on an A4 document to read it in its own Document view: pages and outline on the side, zoom, find, words and reading time. Edit lets you type straight into the pages, with Bold, Italic, Underline, Undo and Redo, saved on this device. Download it as PDF, Word, Markdown or a web page, copy it as formatted or plain text, print it, or open it in Pearl Canvas to ask for changes. Undo all my edits puts it back. An email answer comes as an Email draft with To, Subject (and other subject ideas), the message to edit, Copy, Open in email app and an email file download.
PDF Studio opens files up to 100 MB, up to 3,000 pages at a time. Make smaller shrinks the file (Smallest, Balanced or Clear turn pages into pictures; Keep text keeps the text selectable). Insert a blank page, add initials, a date, a tick or a picture, set the file details, or save pages as PNG or JPG. Ask attaches the finished PDF to the chat with ready questions; it is the only time the file leaves your device. If the page reloads, Pick up where you left off brings your work back.
Data Studio has Try with sample data and Open again for your last file. Ask about this data answers inside the panel, with Copy and Open in chat. Chart cards open full screen and download as PNG or CSV; click a legend item to hide a line. Quizzes remember your progress on this device and end with Review every answer; practice tests can be paused with P.
Code Studio has Find and replace (Ctrl+H), Go to line (Ctrl+G), Search in all files, Format (Shift+Alt+F), Word wrap (Alt+Z), Undo and Redo, and Save now (Ctrl+S). Deleted a file by mistake? Undo shows for 10 seconds, and typing lost when a page closed comes back by itself. Ask AI can change the whole project or only the open file; you apply its changes, except the automatic code check, which applies its own fixes as a new version you can undo. Pearl Canvas More has screen sizes, downloads, Run again, a Console and Compare with the previous version (side by side or code changes). React and Vite projects run as the real page. A project with a server part gets a Server panel with Terminal, Data (SQL), .env (kept only in this browser) and Tests. Press ? in any tool for its keys.
When you ask to change a file made earlier, Pearl AI can show only what changes, in a Changes to <file> card; the full updated file is rebuilt for you and is what Copy, Download, Code Studio and the preview use. In a very long chat, Compacting our conversation tidies the earlier part so you can keep going; What Pearl AI remembers from earlier shows what it kept, and the whole chat stays on screen and in exports.
Not good asks what went wrong and offers a new answer. Earlier answers have Use this version and Back to latest. The Library (Alt+L) filters pinned chats, branches and the last 7 or 30 days, and sorts by date or title. An orange dot after a chat name means it is being answered right now. The Model chip picks Auto or a model your plan has. Used under an answer shows the memories and earlier chats it drew on, each with Forget and Undo. A message waiting while Pearl AI writes has Steer now. Long pastes become a Pasted text card, and + > Take a screenshot attaches a picture of a window on a computer.
Meeting notes runs four steps: everyone agrees, you record (in the room, or an online meeting in another tab, up to 2 hours; the sound is not kept), you check the transcript and rename speakers, and Pearl AI makes the notes with action items you can add to your to-dos. The Interpreter has Swap, Face to face, Big text, Speak translations and Take turns. Ask about a video takes MP4, MOV or WEBM up to 30 minutes (200 MB on a phone, 500 MB on a computer). Edit part of a picture: paint the part with the brush, say what should change and press Make the change. Voice mode opens from the sound-wave button next to Send or + > Voice mode; Hands-free at the top talks inside the chat.
Site names use 3 to 40 small letters, numbers and dashes. Publish as a new site or update one you already have; every update is a version, with Undo last update and Make live for any earlier one. Overview has the link and a QR code. Settings sets the title, the page for addresses that do not exist, who can open the site and the badge. With several own domains, Make main address picks the one visitors get. Forms has Email me new messages; Stats shows 30 days of views and your most viewed pages; Sharing sets the description, icon and the picture shown when the link is shared.
Devices lists everywhere you are signed in, with Sign out all other devices. Add a passkey after a 6-digit email code, then sign in with your fingerprint, face or PIN. Teen mode keeps answers suitable for ages 13 to 17, and turning it off needs a code sent to the account email. App language changes menus and buttons (12 languages). Data & privacy downloads your chats, deletes chats you have not opened for 30 to 365 days (pinned chats are kept) or deletes everything. Files lists every file you uploaded.
Each connector card shows its Recent activity. An Outlook event can get a Teams meeting link. Automations start from ready ideas or on their own when an email (filter by sender or subject), a file, a Google Sheet row, a website form or a secret webhook arrives, and can prepare a reply for you to approve. In Agent mode you can edit the plan, ask before every step, pause, skip, steer and keep it running in the background, and anything that sends, buys, deletes or publishes always waits for your OK (up to 8 steps with a Pearl AI subscription). Deep Research shows each step live, and Refine this report goes deeper, updates it, adds a comparison table or shortens it.
When a job depends on something that only shows later, such as a deploy finishing, a page going live, a price moving or a reply arriving, Pearl AI sets a check-in by itself. It comes back to the same chat at that time (10 minutes to 24 hours later), checks with its tools and reports in a few lines; a notice tells you when it is ready. A card under the answer shows the time and what it will check, with Cancel. Check-ins come with plans that have Background tasks and count in the same daily number; one waits per chat, and at most three run in a row in a chat a day. When the chat makes a reminder clearly needed (a meeting, call, deadline, payment or appointment), it is saved in To-dos on every plan, with Undo on the card.
Open In this chat (the list button at the top of the chat) and choose Overview. Progress shows agent plans and their steps, check-ins, reminders and background tasks. Outputs lists everything made in the chat: artifacts, pages, documents, designs, decks, sheets, files, ZIPs, pictures and research reports. Used in this chat lists the pictures and documents you shared, web sources, memories and earlier chats, connectors and skills. Tap any line to jump to it or open it.
However long a chat gets, questions about it (“my first message”, “your 4th answer”, “the email I gave you”) are answered from the whole saved chat, exactly as you typed it. An address found in a mailbox or a search is never said to be yours. Ask how many chats you worked on today, this week or in total, and which ones, and the answer comes from your real chat list. Every answer is checked against your request: numbers of items, word and page limits, slides, hashtags and language. An answer that stops half-way carries on by itself, one in the wrong form is written again in the right form, and web sources show only when they belong to the answer.
While a document, design or deck is being written you see a card with its title, the pages so far and the part being written. A finished document is never swapped for a one-line note, A4 pages keep their own layout, and the document view fills the panel. Ask for an artifact and you get one designed page at once, with stat cards, icons, tables and real charts; long answers have Make an artifact next to Open as report. Pages of a website show as Open the live website; every website is checked at computer and phone size, and anything to fix is described in plain words with Fix all. Pictures for made-up brands no longer show other organisations’ names, logos or seals.
Stop ends an answer within about a second, even before its first word, and your next message is answered. If stopping takes longer, Send anyway sends your new message at once. Try again shows a spinning icon while it works and says when the result is the same; when trying again cannot help, the message says why and offers the next step. In Code Studio, Run with no file the preview can open names the files and offers Ask AI to make it run.
A greeting or a quick question gets a warm line or two; a real task still gets the complete result, in the language you write in and in exactly the format you ask for. Say remember that and what follows is saved to Memory at once, so the confirmation is always true. Pearl AI never says a website or server is working without a real check, asks once where your data lives before writing a command for it, and reads pasted command or log output like a colleague: what worked, what needs attention and the next step. Asking it to write an email, message or agenda gives you the finished text; Gmail, Outlook and Calendar are used only when you ask to send, draft there or schedule.
Code in answers is read for mistakes and its package names are looked up before you rely on it; the label on the code shows what was checked. Ask whether code works and it says plainly whether it has been run. The automatic fix asks before it changes your files the first time. Ask for only the lines that changed and you get only those, not whole files again. The bar on a code block stays in view while you scroll long code.
In In this chat > Overview, every step of an answer shows live with the apps it looked in, and stays there afterwards. Agent mode now uses the apps you connected (Gmail, Drive, Calendar, Sheets, Microsoft 365, Notion, Slack and your own connectors) in its research, writing and analysis steps, with plans of up to 15 steps; anything it drafts still waits for your Approve.
Send ?? after a few short messages and Pearl AI goes back to your real question; a new question on a different subject is answered as new. Long messages you send fold to a few lines with Show more and Show less. Voice mode asks for the microphone only when you press Start. The preview runs pages apart from your account, so a page there cannot use the microphone or camera: it says so, and you download the page and open it in your browser to try it. The automatic website check runs only on real web pages, not on small pieces of code.
The card under a message with a file you sent opens that file again: PDFs and pictures show in a viewer, text files show their text, and Download gives you your own file. Files sent before 6 November 2026, and files added from Google Drive, open as their text. Only your account can open your files.
When you edit a message, press Try again, or press Retry after a failed send, the pictures and files of that message go with it. In the edit box each one shows as a chip; press its x to leave it out.
Ask for a change to a file you sent (PDF, Word, Excel or text) or to a document or sheet made earlier in the chat, and you get the whole updated document or sheet back, not a piece of it or a script. A PDF comes back as a clean A4 document with your changes; its original layout is not kept. A question about a file, such as “what is the total?”, is answered and nothing is changed. Answers that come from your own connected Google data show no web sources.
A quotation, proposal, invoice, contract or letter for another person or company takes their name and the terms only from your message, your file or the same chat, never from another client’s earlier work. If you have clients on record and did not say who it is for, Pearl AI asks Who is this for? with your clients and Leave the name blank. Otherwise the name is left as a blank line for you to fill in.
Agreements, contracts, NDAs, policies, letters and invoices come as an A4 document in formal English, even when you ask in Hinglish or another language, with the clauses such a document needs (parties, scope, payment and tax terms, notice, confidentiality, governing law and disputes, signatures) and blank lines for anything you have not given. Ask for a 2-page document and it is laid out on 2 pages; any page or slide count in the reply matches the file. Documents, websites and decks use clean line icons instead of emoji, unless you ask for emoji. Have a lawyer check any legal document before you sign it.
Ask for a post, poster, picture or design that is trending, in the latest style, or like a brand or film you name, and Pearl AI first looks up the current style and follows it, without copying anyone’s logo or slogan. A short line with the answer names the style it followed.
When Pearl AI makes a sheet, what it says about the sheet (totals, balance, shares) matches the sheet’s own formulas. Facts that change over time, such as rates and prices, are looked up and given with their date. Asking “did I ask this today?” or “look in my past chats” searches your earlier chats first.
Where these are connected, a follow-up such as “and by country?” or “show the top pages” carries on with the same tool, and “list my properties, sites or channels” lists them all. A Google Meet room is made only when you ask for a room; a calendar booking carries its own Meet link.
The start screen greets you a little differently each time: by the time of day, the day of the week or a welcome back, never the same line twice in a row and in your app language. It uses the name you chose in Settings.
When an answer holds Python or JavaScript, Pearl AI runs it before showing it to you, with the common data libraries (numpy, pandas, matplotlib, scipy, scikit-learn and more). If the code asks for typed input, sample values are used for the check. PHP is checked with PHP 8, and C, C++, TypeScript, JSON, YAML, XML and shell scripts are checked for errors without being run. If the check finds a problem, the code is fixed once before the answer reaches you. Ask does it work? and the answer says what was really run.
Paste your code (or an error message) and ask for a fix in any language, Hinglish included. The answer comes in a fixed shape: What was wrong, the complete fixed code (for a very long file, every changed function and where it goes), What I changed, How to check and Also worth fixing. Your names, structure and file names stay as they are. If you paste a newer version later in the chat, that version is used, and changes that look accidental, such as a field name that changed, are pointed out.
Write still not working, same error or abhi bhi galat aa raha hai in a new chat and Pearl AI continues from your recent chat that holds the code. It names that chat, shows it under Used: earlier chat and gives the complete fix.
Ask Code Studio to rename or remove a function, a method, a page or an input, and the change is checked against every file of the project: calls that still use the old name, routes, links and pages that are still loaded. Anything still to change shows above Apply as Still to update in your project, with an Update those files too button that prepares those changes. Nothing changes in your files until you press Apply.
A website with several pages comes with every page its menu links to, each with real content, before it is shown. When you share your own pages and ask for a change, only the change is made: your pages, links and file names stay exactly as they were.
Pictures with words, realistic photos, ads, posters and logos are made in high quality; simple pictures stay quick. Words you put in quotes are read back from the finished picture, and the picture is made again once if a word is spelt wrong. Ask for an exact size such as 1920x640, or a shape such as 16:9, 9:16, 4:5, an Instagram story or A4, and the picture comes in exactly that size with nothing important cut off; any extra space is filled with a soft background that matches the picture’s edges. Ask for a change to a picture, in any language, and only that part changes. Ask for the real PRSXA or Pearl Organisation logo and the original logo file is placed in a corner. A logo you upload yourself is not placed automatically yet.
Notes in Memory about another client or an earlier project are used for new work only when you name them. Notes about your own business and brand are always used. Email drafts, sheets, calendar events and messages through your connected apps are started only when you ask for one; a how-to question, or the word email inside code you paste, never starts one.
On paid plans, documents, code, plans, research and long answers are written on pearl-6.1, our strongest model. Everyday questions stay on the quick pearl-5-mini, so short answers stay fast. Your usage and credits work the same as before.
With Gmail connected in Settings > Connectors, say check my inbox, what needs my reply or mail kya aaya. Your recent inbox conversations are read and sorted into Needs your reply, Waiting on others, updates and promotions, with one line on what each person wants and how urgent it looks. You are offered drafts for the ones that need a reply; a draft is written only when you ask, and nothing is ever sent without your approval.
When a chat did something you will want again, choose Make one from this chat in the skill picker or in Settings > Skills. The skill’s name, instructions and when to use it are filled in from the chat, in the language you wrote in, and you can add a document from the chat as one of its files. Check it, change anything, and save: the skill is then used by itself whenever it fits.
Long chats now keep more of the conversation in mind before older parts are summed up. Very long documents you add are still read in full and searched as you ask.
Ask for a change to a website made in a chat, such as new button text or a new phone number, and only that part changes. The answer starts with what changed and in which file, and everything else stays exactly as it was. Ask where the other pages are and you are shown where they are: the site’s menu, Open the live website, or the file cards under the answer. Ask for a review or an audit and you first see what was checked and what was found; only those parts are then fixed, with a list of the changes at the end.
When a website is ready, every page is opened at computer and phone size and checked. Layout faults, such as a row of boxes squeezed inside a narrow column, are fixed by themselves once. After any change the ZIP download and the live website hold every page of the site, not only the page that changed. Photos placed in a site are looked at before they are used: a photo of the wrong subject, an old scanned page or a shop sign is replaced, and if no fitting photo is found a plain stand-in is shown in its place.
A pitch deck for investors is written in your voice as the founder and ends with the ask. Figures are never invented: there are no made-up charts or numbers, and if a detail such as the amount you are raising or what the money is for is missing, a short note under the deck tells you exactly what to add.
The search box in the chat list finds words inside your messages and answers too, not only chat titles. Matches show under Found in messages with a short piece of the text; open one and the chat scrolls to it. Changes to a file show as a Changes to … · Applied card, and links whose address contains brackets, such as Wikipedia pages, open in full.
Ask for an email, a letter, an application or a reply and you get a complete, professional piece, sized the way a careful professional would write it. A letter to a school principal, a landlord or a government office covers everything the reader needs to decide; a quick reply to a friend stays a line or two; and if you say keep it short or give a length, that is what you get. Letters get a proper greeting, paragraphs and a real sign-off; reports, ads, social posts and product descriptions get their own layout, with no subject line or signature. Anything only you know, such as an invoice number or a date, appears as a short label like [Invoice Number] with a Fill in line under the piece, so nothing is made up.
A link you paste is read to the end, not just its first part, so the exact wording on a long page, such as a refund policy near the bottom of a terms page, is found and quoted. When an answer searches the web, the best pages it finds are opened and read in full before it answers. News and facts about other companies keep their real names.
Every document made in a chat can be downloaded as a Word file as well as PDF: open the document, choose Download and pick Word (.docx). Headings, lists, tables and photos come across, the document’s top line becomes the page header and page numbers go in the footer. Ask for a document “in a Word file” and the document itself is made, ready to download.
Ask to change a photo you attached and you get the changed picture back. A photo you shared earlier in the chat is used for the next picture or storyboard, so your own product stays in it, and a picture request no longer pulls in results from your connected apps. If making videos is paused for a while, you are told so plainly.
An agent is an automation with a job of its own. Open Automations, choose New automation and turn on Run as an agent. Say who it is (for example “watch competitor prices and tell me when they drop”), when it runs, and what it may do by itself: for sending messages, posting, buying, deleting, sharing personal details, saving to memory, making automations and filling in forms, choose Allow, Ask me or Never. By default it asks before sending or posting and never buys or deletes. Each time it runs it writes a plan in its own chat, works through the steps in the background, waits for your OK where your rules say so, keeps its own notes for next time and logs what it did. Once a week it sends you a report. Each agent counts as one of your automations.
The Your day card on the start screen now has three new parts: Waiting for you lists agent steps that need your OK, with Approve and Review; Done for you today lists what your automations and agents finished, each with a one-line summary and Open; and Coming up today shows what is still to run today.
The row of sources under an answer now shows only the sources the answer actually cites, so each one backs something in the answer. All sources still lists everything that was read. Numbered lists written on one line now show as proper lists.
Ask for a detailed report, proposal, e-book, handbook, white paper, business plan, SOP, manual, policy, guide or thesis (for example “a comprehensive 20-page report on…” or “a detailed business plan for my bakery”). Pearl AI first plans the document, with every section and what it must cover, then writes every section at the same time, each by its own writer, checks them and rewrites any thin section before you see it. The Working panel shows each section as it is written; it usually takes one to three minutes. Reports, proposals and e-books come as designed A4 pages that download as a PDF; plans, SOPs and guides come as one long document that downloads as a Word file. With a Pearl AI subscription a document has up to 10 sections. Short pieces, designs, slides, sheets and websites are made the usual way.
When you mention another chat or something you gave or discussed before (“the code you gave me yesterday”, “the JSON I shared in another chat”, “our Orion project”), Pearl AI looks through your earlier chats first and answers from what is really there, showing which chat it used. If nothing is found it says so and asks; it never makes it up. Questions about your inbox, calendar or Drive are answered only after really looking, when Google is switched on in Settings > Connectors. You can switch earlier-chat search off in Answer options.
For a big job, such as a whole website, a long report or a full plan, start your message with Do this in the background:, type /background, or end the message with “do this in the background”. The message is saved in the chat with the note “Working on this in the background”, and you can close the page or the laptop: the work carries on on our servers, with the same checks as any answer. It usually starts within a minute. A notice at the top right shows Working in the background, then Ready with Open; the answer appears in the same chat and we send you an email. A task that has not started yet can be cancelled from the notice, and if one cannot finish, the notice and the chat say exactly why. Background tasks come with a Pearl AI subscription: up to 20 a day and 2 at the same time, and each uses credits like a normal message. Temporary chats and messages with a picture are answered right away.
Open Apps > Deep Research or type /deepresearch and ask your question. Pearl AI first shows a research plan, the searches it will run, and you can change it. Then it researches on its own for a few minutes, running the planned searches at the same time while you do other things, and saves a full report as a new chat called “Research: …”. A bar shows the progress, Stop ends it at any time, and it can email you when the report is ready. Deep Research comes with a Pearl AI subscription, up to 10 reports a day, and uses credits for its searches and writing.
Say “remind me tomorrow at 10 to call the bank” or “add renew the domain to my to-dos” in any chat and Pearl AI saves it for real, with a card under the answer. At that time the reminder shows in Pearl AI and is emailed to you. Reminders can repeat every day, every weekday, every Monday or every month. Open Apps > To-dos (or type /todos) to add, tick off, snooze, change or delete them, or simply ask “what’s on my to-do list”. The Daily brief, set up with one tap in To-dos, arrives every morning and starts with your to-dos for the day.
Ask for a website with several pages and Pearl AI plans it first, then writes every page in full at the same time, with one shared menu, footer and settings file. “Add more content” or “rebuild the whole site” rebuilds every page; a change to one page stays a small edit. When a website is ready, every page is opened at desktop and phone size and checked: the pages you asked for, links, pictures, sideways scrolling and script errors. The result shows under the answer, and real faults are fixed once automatically unless you tap Not now. PHP websites open in Pearl Canvas too, with working links between pages.
Above each answer, a live Working panel shows what Pearl AI is doing as it happens: what it understood, every web search or page as its own worker running at the same time, the final check and any improvement. When the answer is done it folds into one short line with the time taken and the number of steps.
Ask to be quizzed on any topic or on your own notes and the answer becomes an interactive quiz with instant feedback and a score; the same questions can be practised as flip cards. Press + and choose Upload a recording (MP3, M4A, WAV, MP4 and other common formats, up to 20 MB), or drop the file onto the chat: it is written out word for word, and Pearl AI can then write notes, a summary, decisions and action items from it.
If a feature cannot run for your account, for example because the subscription has ended, the AI credits have run out or a spending cap is reached, Pearl AI says exactly why and shows the button that fixes it. The Voice switch shows “Off” with the reason when the natural voice is not available.
Press Alt+L to see every chat in one place. Search titles and the text of messages, filter by pinned, branches, this week or this month, and pin or delete many chats at once. Any message can start a branch: a new chat with everything up to that point, while the original chat stays as it was.
Press Alt+C for projects with many files: a code editor, Run to preview in Pearl Canvas, and Ask AI to change files, shown as changes you apply yourself (the automatic code check applies its own fixes, which you can undo). Projects are kept in your browser.
Data Studio opens CSV and Excel files up to 50 MB (200,000 rows): sort, filter, see each column's profile, draw charts, make summary tables, clean up, export, and ask questions about the data. PDF Studio (Alt+P) merges, splits, reorders, turns and removes pages, signs, fills in forms, adds watermarks and page numbers, turns pages into pictures and copies the text. Both work in your browser, so your files stay on your device.
Video Studio (Apps > Video Studio, or type /video) makes videos from your words, from your own picture, or from a talking presenter. Choose Draft (a quick preview), Standard (HD) or Premium (our best quality, on Max), a wide, tall or square shape, and a length: One clip, or a Long video made as shots that each carry on from the one before and are joined into one video by themselves. Then join and trim clips and add voiceover, music, titles, captions and your logo, or build a full trailer. Videos are kept for 30 days. Every plan has free Draft seconds each month: Free 10 seconds, Plus 15 seconds, Pro 1 minute and Max 5 minutes. After them, and for every Standard, Premium or presenter video, paid plans use extra credits, taken only when the video is made; Free can upgrade. The longest clip is 5 seconds on Free, 10 on Plus and 15 on Pro and Max (Premium up to 30), and a Long video can be up to 30 seconds on Plus, 1 minute on Pro and 3 minutes on Max.
CAD and 3D Studio (Apps > CAD and 3D Studio, or type /cad) turns a description or your sizes into a live drawing and a 3D model: 148 shapes across 15 groups — mechanical parts, pipes, electronics, automotive, architecture, furniture, floor plans and more — or describe any object in words and the studio picks the closest shape. Type the sizes, choose a colour, and see a live 2D drawing and a 3D model you can turn and zoom. Build from parts lets you assemble or cut shapes together. Floor plans support up to 24 rooms with doors, windows and archways, measured inside the walls; describe a whole home, office or shop in words and a furnished plan is drawn for you. Each saved design gives you DXF, SVG, PDF, STL, OBJ and GLB files, kept for 60 days, and no credits are used. Saved designs a day: Free 3, Plus 10, Pro 25, Max 60. It is made for simple and medium parts and plans, not for engineering-grade CAD: always check the sizes before you make or build anything.
While Pearl AI is writing, the browser tab icon shows a softly pulsing orange dot; it turns green when the answer is ready and red if a reply could not finish. If the answer finishes while you are in another tab, the tab title starts with "Answer ready", or "Needs your OK" when Agent mode is waiting for you. Desktop notifications are optional and you can turn them off in Settings.
Publish from Pearl Canvas or Code Studio and your site gets a free https address at once. The Publish window has Overview, Versions (the last 20, put any one back), Settings (anyone with the link, or a password), Forms (messages from your site’s forms, also sent to you by email), Stats, SEO and Domain. In Domain, type your domain (for example www.yourbusiness.com), add the two records shown where you bought it (there is step-by-step help for GoDaddy, Namecheap, Hostinger, Cloudflare, BigRock and others) and press I have added them. Pearl AI checks them and connects your domain with a secure https address for you, usually within an hour. With a Pearl AI subscription you can connect up to 5 domains. Settings > Publishing lists every site and domain with Manage, Copy, Open and, under the three dots, Unpublish, Change and Remove. Published sites show a small “Powered by Pearl AI” badge with a Report link, and it stays on.
Press the sound-wave button next to Send (or + > Voice mode), and then Start. Talk normally: a short pause sends what you said, Pearl AI answers out loud and shows the whole answer with the spoken word underlined, then listens again. Interrupt cuts in (on a computer you can also just start talking), the microphone pauses and the red button ends the call. The sliders button opens voice settings: natural voice (uses AI credits), the voice, the speed and the language you speak (the main language of almost every country). Camera and Screen (computers) let Pearl AI see what you show: one fresh picture goes with each thing you say.
In the + menu: Meeting notes records a meeting in the room or in an online meeting tab (everyone must agree first) and gives a summary, decisions, action items you can turn into to-dos, open questions and key moments, with a Word download; the sound itself is never kept. Interpreter lets two people who speak different languages talk face to face, translated and spoken live. Ask about a video takes a video of up to 30 minutes, watches and listens to it on your device (the file is never uploaded) and answers with times. Edit part of a picture lets you paint over an area and say what should change.
With a Pearl AI subscription, Settings > Connectors connects Microsoft 365 (Outlook email and calendar, OneDrive), Notion and Slack in steps that turn green. Pearl AI finds what you need and drafts emails, events, files, pages and messages as cards you edit and Approve; nothing is sent before that. The WhatsApp card links your WhatsApp number in three steps, then you can chat with Pearl AI on WhatsApp with voice notes and pictures. The same page has the Word, Excel and Outlook add-ins and the browser extension (each sign-in is approved with a short code, and every browser action asks you first). Automations can now start when something happens: a new email, file, sheet row, website form or webhook.
The Team button in the sidebar opens Group chats (your team and Pearl AI in one chat; each person’s questions use their own account) and Shared docs (documents you edit together live, with comments, Ask AI suggestions to accept or reject, and version history). Spreadsheets work the same way: press Share live on a spreadsheet in a chat, add people from your team or turn on the link, and everyone edits the same cells at the same time. You see who is on which cell, every change shows within a second, and you can comment on a cell with @mentions, replies and Resolve. If two people change the same cell at the same moment, nothing is lost: you choose Use mine or Keep theirs. Versions keep earlier copies to restore or download as Excel; viewers and commenters cannot change cells. In a project, project memory keeps notes every chat in that project remembers; say “remember in this project: …”. A Study hub with flashcards, trip plans with a map and shopping comparisons with sources are built in, and a slide deck can be saved to Google Slides.
Apps with a server part in Node.js, Python or PHP run live in the preview, with a small database, a test sign-in and a Server panel. Settings > Install app shows how to install Pearl AI on your phone, tablet or computer, with notifications. Add a passkey in Settings > Security & sign-in, switch on Teen mode in Settings > Age & safety, and find every file you shared in Settings > Files. Web answers keep numbered sources with the full list, and with a subscription you can choose the model for your answers.
In-depth answers work on every Pearl AI account. Big requests, such as reports, plans, proposals, long articles, guides, comparisons, several questions at once, code, or anything you ask for “in detail”, are planned first (Planning the answer), given about twice the room of a normal answer, continued by themselves up to 2 times so nothing is cut off, and checked so every part you asked for is there (Checking it is complete; anything missing is added under the answer). Quick questions stay quick. With a Pearl AI subscription these answers are written on the stronger model and use a little more credit.
The code check works on every Pearl AI account. After every AI change, Code Studio runs your app in the preview, looks for script errors, console errors, missing files, a blank page and server errors, fixes them by itself and checks again. A card shows the steps turning green, ending in Works only when the last check found nothing; otherwise it lists each problem with a link to the file and line, and offers Try fixing again, Undo the fixes and Check again. It never changes how your design looks by itself: something that only looks wrong in the check (a button that fades in, or a phone menu button hidden on a computer) is shown as a note to look at, and a fix that would take out a stylesheet, a script, a file or a large part of a file waits for your OK (Keep my files or Apply the fix). With a Pearl AI subscription it runs up to 3 automatic rounds, without one 1 round. A complete app in a chat answer gets a quiet check and a Fix this code chip when it shows an error. Ask for the full or updated code and you always get the whole file with every change, never only the part that changed.
Advanced coding is on every Pearl AI account: code tasks are answered as complete, working files with every import and edge case handled, continued if long and checked so nothing such as “the rest of the code” is left out, and code reviews list real problems by severity with the exact fix.
Meta Ads and Google Ads need a Pearl AI subscription. In Settings > Connectors > Ads & social, connect Meta Ads (Facebook and Instagram ads) or Google Ads, sign in and tick the ad accounts Pearl AI may use. Then ask how your ads did: Pearl AI answers with real figures from the ad account in its own currency (spend, reach, clicks, results, cost per result, return on ad spend), compared with the period before, and gives advice. Pausing, resuming, budget changes and new campaigns, ad sets or ads are prepared as cards and happen only after you press Approve; one budget step stays between 0.2 and 3 times the current budget, and new items are always created paused.
Facebook Page & Instagram posting needs a Pearl AI subscription too. It connects your Facebook Pages and the Instagram accounts linked to them: Pearl AI reads recent posts and results and gives advice, and posts only after you approve a card with a real preview (Facebook: text, a link or 1 to 10 pictures, now or scheduled up to 30 days ahead; Instagram: one picture or a carousel of 2 to 10, published straight away). It can also reply to or hide comments after you approve. Meta and Google are still reviewing Pearl AI’s access, so connecting may not finish for every account until their reviews are done.
In Settings > Connectors > WhatsApp > Your business number, connect your own existing WhatsApp Business number and Pearl AI answers your customers there from the setup you write (what you do, FAQs, opening hours, prices). It only replies to customers who wrote first, inside WhatsApp’s 24-hour window, and never messages anyone first. Your team members can add their own numbers on your account (Team numbers). The live inbox shows every chat with search and filters: Take over stops Pearl AI for that chat and Hand back to AI starts it again; Pearl AI Chat’s Conversations tab shows the same inbox under WhatsApp. Replies stay inside Meta’s 1,000 free replies a month per number unless you switch on paid replies, which Meta bills to your own WhatsApp account. If credits run out, answering pauses and your customers’ messages are still saved. This is part of a Pearl AI plan with AI credits, because the answers are paid from the credits, and it includes 1 business number; a Pearl AI plan that counts monthly usage instead does not include it, and the connect step says so. Connecting opens once Meta finishes verifying Pearl Organisation.
Click any picture in a chat to zoom with + and −, Fit, 100%, the mouse wheel, a double-click or two fingers, and drag to move around; pictures you share are kept as a clear private copy. In voice mode, about a second after you ask, Pearl AI says a short line such as “Sure, one second” while it looks things up, so there is never a silent wait. Saying “Hey Pearl AI” with your question into the message-box microphone opens voice mode by itself, and on computers you can switch on Listen for “Hey Pearl AI” in the voice settings.
Typos are read by meaning, and “I meant …” makes Pearl AI re-read the chat and answer what you really asked. Words such as “only the text”, “no code” or “not a website” are followed exactly. Text you want in a picture is kept word for word and fitted inside it. Ask why something is not available and Pearl AI tells you the real reason from your account and settings at that moment, and where to fix it (for example Settings > Billing or Settings > Connectors).
With a Pearl AI plan with AI credits, voice mode runs as a Live call. Pearl AI starts answering about a second after you stop talking, you can talk over the answer at any time (even on a phone’s loudspeaker), and live captions show both sides (your own words are written out as you speak and saved in the chat). A green badge shows Live and the minutes left, and the call screen has Interrupt, mute, camera and End; with the camera or screen on, each turn uses a fresh picture. During the call Pearl AI can check live facts such as weather, prices and scores, search the web and read your saved notes. Anything that would send, post, book or create something is drafted in the chat as a card that waits for your Approve. Both sides of the call are saved in the chat, and saying “bye” ends the call after a short goodbye.
Live call uses AI credits, with up to 240 Live call minutes a month that start again on the 1st; one call lasts up to 30 minutes. Settings > Usage shows your credits. When the minutes or credits run out, or the connection drops twice, the standard voice carries on in the same window without hanging up, and a short note says why. In voice mode the sliders button opens the voice settings: the Live call switch (on by default), Live call voice with 10 voices, Speed (it also changes the voice during a call), and the minutes left this month. While Pearl AI checks something during a call, it says a short line, so there is never silence. The Live call voice model is pearl-voice-live.
The standard voice now works smoothly on phones too: on Android and iPhone a short pause no longer cuts your question, Pearl AI never hears itself through the loudspeaker, answers play to the end (sound that pauses after a call or a notification on iPhone comes back by itself), and the first words come sooner. On Samsung Internet, and in the iPhone home-screen app when the phone does not allow the browser’s own speech typing, Pearl AI writes out what you say itself.
Messages, live web searches, created images, automation runs, Deep Research reports, background tasks, scanned PDF pages, written-out recordings and the natural HD voice use AI credits. Think deeper answers use more credits than quick ones because they do more work, and when Pearl AI runs code, the step where it reads the results and writes the final answer counts as a message. Uploading documents, exporting, sharing, editing, Data Studio and PDF Studio are free (asking Pearl AI about your data counts as a message).
Press + > Upload a document and pick several files at once, or drop them into the chat. One message carries up to 5 documents on a Pearl AI plan (2 on the free trial), up to 50 MB together. Each file shows its own chip with a remove button while it uploads, and if one cannot be added (a kind Pearl AI does not read, too big, or one file over the number for one message) the reason shows right under that chip, so you can remove it and send the rest. Pearl AI reads every file of the message, and each file still counts in the documents a day.
On PRSXA™ the number of documents in one message follows the plan: Free 2, Plus 5, Pro 5, Max 5, with up to 10 MB together on Free and 100 MB on Max.
How much of the earlier conversation Pearl AI keeps in mind in a long chat. With a Pearl AI plan it is about the last 100 messages of a chat (up to about 27,000 words); on the free trial about 40. Older parts of a very long chat are kept as a short summary, so if something from much earlier seems missed, ask Pearl AI to recap it. Settings > Usage shows the number for your account.
On PRSXA it grows with the plan: Free 60, Plus 120, Pro 250, Max 400 messages of a chat.
A skill is instructions you use again and again, saved once: for example “Weekly report” with your layout, or “Brand voice” with your tone and words to avoid. Each skill has a name, what it is for, the instructions, up to 3 files (such as a brand guide or a price list) and trigger words. Your own skills come with a Pearl AI subscription, up to 50 skills.
On PRSXA your own skills are part of Plus, Pro and Max: Plus 10, Pro 50, Max 200 skills. When the number is reached, the message under the Save button says so and names the plan that keeps more.
Anyone can try PRSXA at prsxa.com/chat without an account: 10 questions a day, answered on pearl-5-lite with no web search, pictures, files or memory. The quickest way in is the question box at the top of the prsxa.com home page: type a question and the chat opens with it already being answered. When the questions are used, a card offers a free account; after signing up (or signing in) that chat moves into the account by itself. Pearl AI itself has a 5-minute free trial for new accounts.
Connect your Google account and Pearl AI can find your emails and files when you ask, and write emails and meetings for you. Open Settings > Connectors (Alt+,). Each part has its own switch, and every switch starts off:
An email or meeting appears as a card under the answer. You can change the recipients, subject, text, time and guests on the card, then press Approve to send or add it, or Discard. If something on the card needs fixing, the message shows right under that field. Pearl AI never says an email was sent or a meeting was booked before you approve it.
A Google Sheets change appears as a table that shows every row and cell that will be written. Press Approve and save (or Approve and create for a new sheet), or Discard. To change the data, ask in the chat and a new card is made. Formulas that could fetch data from the internet are saved as plain text.
Pearl AI only writes to addresses that you typed, that appear earlier in the chat, or that it found in your own emails, calendar or files. It never guesses an address, and example addresses are always refused. To answer someone who wrote to you, say "reply to this email": Pearl AI takes the sender's address from your emails, and asks you only if it can't find it. You can always type any address on the card yourself.
Press Connect Google and choose your account on Google's own sign-in page. Pearl AI asks only for the parts you switch on, and you can add more later. Personal Google accounts can use Google Drive now. The other parts (Gmail, Google Calendar, Google Analytics, Search Console, YouTube and Google Sheets) open to personal accounts once Google finishes its review of the app; company accounts set up by Pearl Organisation have every part.
PEARLian accounts (@pearlorganisation.com) are connected through the company's Google Workspace, so there is no sign-in step. They have Read Gmail, Send email and Google Drive, but not Google Calendar: meetings are set up through Quick Pearl Meet and Pearl Meet in the ERP (open Pearl Meet).
Besides Google, Pearl AI can connect to other apps and to your own systems, as long as they offer an MCP server. MCP is a common standard that lets AI assistants work with other software; many software companies publish an "MCP server address" in their help pages, and your own developers can make one for your system. Once connected, Pearl AI reads from that app or acts in it when you ask, for example "which tasks are due this week in my tracker?" or "add a note with these meeting points". Custom connectors are part of a Pearl AI subscription (in PRSXA: Plus, Pro and Max).
Advanced settings are only for servers that need them: an Access key header and Access key when the server gives you a key instead of a sign-in page (the key is stored encrypted and never shown again), or an App ID and App secret when the server needs its own app registered first; the form then shows the return address to use in that app's developer settings.
Each connector shows its status (Connected, Off, Sign in again or Not reachable) and has an on/off switch, Refresh tools, Sign out and Remove. You can add up to 10 custom connectors; Pearl AI and PRSXA keep separate lists. Each answer can use connectors up to 8 times.
Many tools: when your connectors offer more than 30 tools, each message uses the ones that fit it best, and Pearl AI loads any other tool the moment it needs it, so nothing is left out.
You can see where an answer came from: every connector step shows in the Working panel above the answer ("Checking Notion", then "Checked Notion"), and an answer that used your apps shows From your connectors with their names under it, next to any web sources; point at a name to see which tools were used. Open as report and the PDF and Word downloads list them too. When your message names a connected app ("using Notion, ..."), Pearl AI goes to that app first.
This works for every kind of question, including questions about the latest facts and Research mode: the app you named answers first, and Pearl AI searches the web only for what the app does not have. The name works with or without spaces, for example "DeepWiki", "Deep Wiki" or "deep-wiki".
Research: research rounds in a chat can use your connectors next to web search. Deep Research reports also look in your connectors, using only tools that only read and are set to Allow; nothing is ever changed in your apps by a report. Facts from your apps are named in the report, for example "in your Notion".
You can ask Pearl AI about your code from inside your code editor. Open the Connector directory (or Settings > Connectors > Code editors) and press Connect on your editor. PRSXA and Pearl AI each have their own connections.
Each editor shows Connected once it is in use, or Waiting until then. Manage > Disconnect removes it at once, and you can connect up to 10 editors. Treat a copied key like a password: if it is ever shared, disconnect that editor and connect it again.
Give Pearl AI your own information once, so answers use your real facts instead of general ones: a business adds its price list, policies and FAQs; a freelancer adds their rate card and terms; a student adds their CV or study notes. Open Settings > Knowledge (Alt+,). If you have a team, everyone in it uses the same knowledge.
Plans: up to 25 items on PRSXA Plus, 100 on Pro or a Pearl AI subscription, 300 on Max. Team members use the owner's knowledge whatever their own plan.
Settings > Team and audit is where anyone builds a team. Invite people by email (up to 5 on the PRSXA Free plan, 10 on Plus, 25 on Pro and 50 on Max, or 25 with a Pearl AI subscription; moving to a smaller plan never removes anyone, it only stops new invites) as Viewer (uses your knowledge and can see your PEARL PRODUCTS account) or Admin (can also add and remove knowledge, set up your products, buy plans and top up credits). Resend or cancel invites, change roles and remove people in the same place. Invited people accept there or from the invitation email; they keep their own account, chats and plan, and can leave a team at any time. Nobody in a team can see your chats, memory or Google, unless you share a chat with your team: open the chat, tap Share at the top, then Share with your team. Members read a copy under Settings > Team and audit > Chats shared with your team (never uploaded photos or files, and temporary chats can't be shared), can tap Continue in my own chat for a private copy, and lose access at once if they leave or are removed. Update adds your newest messages, Stop sharing ends it, deleting the chat ends it too, and every share shows in the audit. As the owner you then decide how your members use Pearl AI, and members see the rules that apply to them:
When someone is in more than one team, the strictest rule applies. Rules take effect from the member's next message.
Press Alt+, to open Settings, or press the Usage button at the top of the chat (in the More menu on a phone) to go straight to your usage. Settings has about 30 sections in groups, and the search box at the top finds any setting by name. Your choices are saved to your account, so they follow you to every device.
A bar shows how much of your monthly credits you have used, in dollars, and the date they reset. Below it you see your extra credits, which are used only after the monthly credits run out and do not expire at the monthly reset, with Buy more and a link to auto-reload. If your monthly credits are used up, a note at the top tells you, and says whether extra credits are now being used. Press the refresh button for the latest numbers. On a new account, Usage shows the time left in your free trial. Everything from the earlier credits page is still there under Credits details and options.
Set the most your Pearl AI plan may spend in a calendar month, in US dollars. On the Monthly spend limit card press Manage, type the amount (from $1 to $100,000) and press Save limit; Remove limit takes it away. The card shows what was spent this month, your limit and a bar that turns amber at 80% and red when the limit is reached. The limit covers the chat page, the API and assistants together, and it is the same limit as the monthly spending cap in API & developer. When it is reached, Pearl AI pauses until the 1st of next month, or until you raise or remove the limit; your credits stay where they are. Only the person who owns the plan can change it. In PRSXA (Plus, Pro and Max) the limit is in credits and covers only extra credits used after your plan's allowance runs out.
How many chats (messages), pictures, files read, code runs, Deep Research reports, background tasks, automation runs, API calls and assistant calls you had since Monday, in your own time zone, each with a bar for its share of the week. Press the round arrow for the latest numbers.
Everything about paying in one place: your plan and the date it renews or ends, an Auto-renew switch, and Adjust plan. Payment method shows the saved card (brand, last four digits and expiry) with Update: one card pays renewals and top-ups for every Pearl product. Extra credits shows your balance with Buy more. Auto-reload adds credits when your balance falls below an amount you choose, with a monthly limit so it never spends more than you allow; you can edit it or turn it off at any time. Every payment is listed below with its receipt to download. If your Pearl AI plan is billed through Pearl Client Workspace (PCW), Adjust plan and Buy more open it in a new tab.
Choose what Pearl AI does for you:
Changes you make from the options in the message box are saved here too.
Two lists of up to 10 websites each. Search these first: pages from these websites are looked at first and listed first as sources; other websites are still used. Never use these: pages from these websites and their sub-sites are never used or shown as sources. They apply to every web search Pearl AI runs for you, in chats, Deep Research and automations, and they never add searches.
Besides theme, text size, width and density, Code appearance sets the colours of code in answers, one choice for light mode (Pearl, GitHub Light, One Light, Solarized Light) and one for dark mode (Pearl Dark, GitHub Dark, Monokai, Dracula, Nord), each with a small live preview. You can also type the name of a coding font installed on your device, such as JetBrains Mono. Interface font changes the font of menus, the sidebar and answers: Standard, Serif (Literata, a book font made for long reading on screens) or Italic (Lora Italic). Code keeps its own font.
Reflect shows how you have used Pearl AI over the past week, month or three months, in your own time zone: conversations, messages sent, about how much time you spent and on how many days, a day-by-day chart, your most active day and your busiest hour. Temporary chats are never counted. Once a day Pearl AI also writes a short summary of the main themes, using only your chat names, as shown in your chat list, and these numbers. It never reads the messages themselves for this, and it leaves out anything about health, money, relationships, religion, politics or legal matters. Switch off Write a summary for me to see the numbers only. Only you can see Reflect.
Break reminders give a gentle nudge after 30 or 45 minutes, 1 hour, 1.5 hours or 2 hours of chatting without a pause, or never. You can snooze a reminder for 5, 10 or 15 minutes. A pause of 15 minutes or more counts as a break, on any of your tabs, and Start fresh resets the timer. Quiet hours lets you pick the days and a time from and until. A period that runs past midnight belongs to the day it starts. During quiet hours Pearl AI shows a calm reminder and stays silent, with no sounds or alerts, and you can always carry on if you need to.
On a computer, when the sidebar is closed, move the pointer over the sidebar button or the left edge of the screen and it slides open over the chat, so you can pick a chat without changing your layout. It closes again when the pointer leaves, when you press Esc or when you pick a chat. Press the pin button to keep it open.
The quickest way to see what Pearl AI can do, before creating an account or writing any code, is the playground. Open the Playground page from the products site and start typing a question — you'll get a reply within a few seconds.
The playground exists so you can judge, in your own words and on your own topics, whether Pearl AI's answers are useful to you before committing to anything. It isn't meant to be used as a permanent tool — for ongoing use, create an account and either use the full chat page or connect to the API.
Authentication simply means "proving who you are" before the system lets you in. Pearl AI uses two different methods depending on what you're doing:
| Method | Used for | How long it lasts |
|---|---|---|
| Signing in with your email & password | Using the website — your dashboard, and the built-in chat page, in a browser | Stays signed in while you're active; you can sign out any time |
| An API key | Connecting your own website, app, or script directly to Pearl AI | Works until you revoke it or it reaches the end date you set |
To get an API key: sign in, open Configure Pearl AI → API access, and choose + Create key. You can create as many keys as you need; each one is shown only once, can be limited to chosen assistants and IP addresses, can have an end date, and shows its own usage. Creating a key needs a one-time code and sends a security email. A key is a long string of letters and numbers that acts like a password for your account's AI access — anyone who has it can use your account and your credits, so treat it with the same care as a password.
This part is written for a developer connecting Pearl AI to their own software. If that's not you, feel free to skip ahead to Credits & usage.
Pearl AI's chat endpoint ("endpoint" = a specific web address your code sends a request to) accepts a short conversation and sends back a reply, in the same general shape used by most AI chat services — so if a developer has connected to one of these before, this will look familiar.
POST https://products.pearlorganisation.in/api/v1/chat/completionsAuthorization: Bearer YOUR_API_KEYapplication/jsoncurl https://products.pearlorganisation.in/api/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"messages": [
{ "role": "system", "content": "You are a concise, helpful assistant." },
{ "role": "user", "content": "Summarize what Pearl AI does in one sentence." }
]
}'
// Node.js
const response = await fetch("https://products.pearlorganisation.in/api/v1/chat/completions", {
method: "POST",
headers: {
"Authorization": "Bearer YOUR_API_KEY",
"Content-Type": "application/json"
},
body: JSON.stringify({
messages: [
{ role: "system", content: "You are a concise, helpful assistant." },
{ role: "user", content: "Summarize what Pearl AI does in one sentence." }
]
})
});
const data = await response.json();
console.log(data.choices[0].message.content);
# Python
import requests
response = requests.post(
"https://products.pearlorganisation.in/api/v1/chat/completions",
headers={
"Authorization": "Bearer YOUR_API_KEY",
"Content-Type": "application/json",
},
json={
"messages": [
{"role": "system", "content": "You are a concise, helpful assistant."},
{"role": "user", "content": "Summarize what Pearl AI does in one sentence."},
]
},
)
print(response.json()["choices"][0]["message"]["content"])
These are the extra settings a developer can include alongside the conversation itself:
| Option | Required? | In plain terms |
|---|---|---|
messages | Yes | The conversation so far — a list of who said what (system sets the assistant's behavior, user is the visitor, assistant is a previous AI reply). |
model | No | Which model answers: pearl-4-mini (fast, low-cost, used by default) or pearl-4 (higher accuracy). |
temperature | No | How predictable vs. varied the reply should be. Low = focused and consistent; high = more creative and varied. |
response_format | No | Ask for the answer as JSON, optionally in an exact shape. See Advanced options. |
stream | No | true sends the answer word by word as it is written. |
thread_id | No | "new" or a thread ID, so Pearl keeps the conversation and you only send the newest message. |
async | No | true answers in the background; collect the result later or by webhook (assistants only). |
conversation_id | No | Your own ID for a conversation, so "Ask before acting" can wait for the user's yes across requests. Not needed with threads. |
A successful reply includes the AI's answer, plus a short usage summary showing how much of your credit balance that one request used.
Every request uses your API key (it starts with pai_) in the Authorization: Bearer header. The base address is https://products.pearlorganisation.in/api.
| Endpoint | What it does |
|---|---|
POST /v1/chat/completions | Send a conversation, get a reply. |
POST /v1/assistants/{assistant_id}/chat/completions | Send a conversation to one of your assistants, with its saved instructions and your variables. |
GET /v1/assistants | List your assistants and whether each is switched on. |
GET /v1/threads/{thread_id}/messages | Read a conversation Pearl kept for you (DELETE /v1/threads/{thread_id} removes it). |
GET /v1/jobs/{job_id} | The result of a background answer. |
pearl-4-mini is fast and low-cost, and is used when you don't name a model. pearl-4 gives higher accuracy and better reasoning. Choose one with the model field in your request.
Each API key can make up to 60 requests a minute. Every reply shows the credits that request used and the credits you have left.
Each assistant's Code button gives ready-to-copy examples (Node.js, Python, cURL) and a Postman collection. To let AI apps such as Claude, Cursor or VS Code read your plan, open Connect your AI tools on the API Access tab: it gives them read-only access with a key you control.
Open the dashboard with Configure Pearl AI: on My Subscriptions (PEARL PRODUCTS site) or My Pearl Products (Pearl Client Workspace). The top card, Pearl AI workspace, shows your credits left, requests and credits used (Pearl Client Workspace counts this month; the PEARL PRODUCTS site counts the last 30 days), with Open Pearl AI and Activity & usage.
The tabs are, in order: Setup, Assistants, API access, Credits and Privacy & history (shortened to Privacy on narrow screens). Follow them in that order the first time:
Under the guide line you'll find the How Pearl AI answers box, the Start from a template card, and the Default AI behaviour card (marked Default brain):
Templates are ready-trained roles for every major industry, team and personal need, plus task templates that return clean JSON for your apps (for example an invoice reader, lead scorer or ticket router). There are two template cards:
Both show a few quick picks, See all and Browse templates.
[not provided yet] are the gaps to fill on its Knowledge panel.The Assistants tab shows N of M active and, while you have a free place, + New assistant. When every place in your plan is in use, an amber banner says You're using all … assistant(s) in your plan with Buy another assistant ($20, valid until your plan renews).
{{customer_name}} anywhere and a table appears with Required, Default value and What it is. Names start with a letter and use letters, numbers and _; up to 20 per assistant.asst_…): click it to copy.Under a new assistant, Finish setting up · N of 4 done guides you. The next step is outlined in blue; finished steps turn green.
Press Hide steps to fold them away; a Show setup steps button stays on the card with the count, so you can bring them back at any time. Steps 3 and 4, and hiding, are remembered in that browser, so another computer may show them again. When all four are done, the steps disappear and the card shows Ready.
Press Design on an assistant's card. The window shows a live preview (switch Phone or Wide; on tablets and phones switch Edit and Preview) and these tabs:
#2354D6), Light or Dark theme and corner roundness.Save design appears once you change something. Before saving, a box asks Save this chat design? Every chat window for this assistant shows it right away. Undo last save brings the current one back. Press Save design to go ahead. Undo last save puts back the design saved before, and Back to template design / Back to standard design replaces your own design (both ask first). Closing with unsaved changes asks Close without saving your changes? with Keep editing.
Open More → Answer format. Choose how answers come back:
For each field, give a name (letters, numbers and _, starting with a letter), a type (Text, Number, Whole number, Yes / No, Choice, List or Group of fields), an optional "What goes here" hint, and whether it can be empty. Up to 60 fields, 4 levels deep, and 50 allowed values for a choice.
Each key shows when it was last used, a 14-day chart and any failed requests. Revoke asks first and stops the key at once. The Connect your AI tools card lets AI apps read (never change) this plan's data with a connection key you create, limit and can disconnect at any time.
An assistant is a saved set of instructions for one job, for example replying to support tickets, summarising leads, or checking quality. Each assistant has its own address, and you call it with your normal API key, so your code only has to send the conversation.
The monthly plan includes 1 assistant and the 12-month plan includes 2. You can add up to 50 more for $20 each; they last until your plan's current end date and are not carried over. See Plans & pricing.
{{customer_name}} or {{plan}}. Each variable you type is listed under the instructions automatically.curl https://products.pearlorganisation.in/api/v1/assistants/YOUR_ASSISTANT_ID/chat/completions \
-H "Authorization: Bearer pai_YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"messages": [ { "role": "user", "content": "Hello" } ],
"variables": { "customer_name": "Asha", "plan": "Pro" }
}'
The values in variables are filled into the instructions before the reply is written. If a variable the assistant needs is missing, the request is refused with a message naming it.
Common mistake: the assistant ID (it starts with asst_) goes in the address, while your API key (it starts with pai_) goes in the header. Sending the assistant ID as the key returns an error saying so.
You can switch an assistant off without deleting it; a switched-off assistant replies that it is unavailable. You can have as many switched on as your plan allows. To switch on another, switch one off or add an extra assistant.
Every reply from an assistant is paid for from your Pearl AI credits, the same as a normal request.
Each assistant card in Pearl AI → Configure → Assistants has buttons that make the assistant more useful. Everything you set up here is used automatically when your app calls the assistant through the API.
Tools let the assistant look things up and take actions in your own systems while it answers, for example checking an order's status or creating a booking. Add your own web address as a tool, or connect an MCP server (many services such as Notion offer one; some ask you to sign in once). For tools that change something, you can switch on Ask before acting: the assistant then asks the user to confirm before it goes ahead. Every tool call is logged, and the answer lists the tools it used in tool_calls.
Knowledge is what the assistant should know about your business. Add notes, exact questions and answers, up to 5 documents (PDF, Word or text, up to 5 MB each) and your website, which Pearl reads and keeps up to date. For each question, Pearl adds only the parts of this knowledge that match it, so answers stay focused and use fewer credits.
Every saved change to an assistant is kept in its Versions (under More), so you can compare it with what's live or restore an earlier version. Test sets let you save questions and re-run them after a change to check the answers still look right.
Try it opens a chat with the assistant right on the page, working exactly as your app would: the same instructions, variables, knowledge and tools. Each answer shows the knowledge and tools it used, its size and cost, and API request gives you the matching request to copy into your code. Answers here use your credits like API calls and appear in the Activity log as playground tests.
Four optional settings for apps that need more than a simple question and answer. They work with /v1/chat/completions and with your assistants (background answers are for assistants only).
Ask for the answer as data your code can read straight away. Use {"type": "json_object"} for any JSON, or give a schema to get exactly the fields you need. The answer is in choices[0].message.content as text and, ready to use, in choices[0].message.parsed.
curl https://products.pearlorganisation.in/api/v1/assistants/YOUR_ASSISTANT_ID/chat/completions \
-H "Authorization: Bearer pai_YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"messages": [ { "role": "user", "content": "I want to cancel order A-1001, it arrived broken." } ],
"response_format": {
"type": "json_schema",
"json_schema": {
"name": "ticket",
"strict": true,
"schema": {
"type": "object",
"properties": {
"intent": { "type": "string", "enum": ["cancel", "refund", "question", "other"] },
"order_id": { "type": "string" },
"urgent": { "type": "boolean" }
},
"required": ["intent", "order_id", "urgent"],
"additionalProperties": false
}
}
}
}'
If the schema itself cannot be used, the request is refused with 422 and a message saying what to change.
Add "stream": true and the answer arrives word by word as it is written, so your users see it straight away. The reply is a stream of data: lines (server-sent events) in the standard chat-completions streaming format: each piece has its text in choices[0].delta.content, the last piece also carries usage and credits, and the stream ends with data: [DONE]. When the assistant uses tools, the tools run first and the answer then follows in pieces.
const res = await fetch("https://products.pearlorganisation.in/api/v1/assistants/YOUR_ASSISTANT_ID/chat/completions", {
method: "POST",
headers: { "Authorization": "Bearer pai_YOUR_API_KEY", "Content-Type": "application/json" },
body: JSON.stringify({ stream: true, messages: [{ role: "user", content: "Tell me about your plans" }] })
});
const reader = res.body.getReader(), decoder = new TextDecoder();
let buffer = "";
while (true) {
const { value, done } = await reader.read();
if (done) break;
buffer += decoder.decode(value, { stream: true });
const events = buffer.split("\n\n"); buffer = events.pop();
for (const e of events) {
if (!e.startsWith("data: ") || e === "data: [DONE]") continue;
const piece = JSON.parse(e.slice(6));
if (piece.error) { console.error(piece.error.message); continue; }
process.stdout.write(piece.choices[0].delta.content || "");
}
}
Normally your app sends the whole conversation with every request. With a thread, Pearl remembers it for you. Send "thread_id": "new" with the first message; the answer includes a thread_id (it starts with thr_). From then on send only the newest message with that thread_id. Pearl uses the last 40 messages of the thread for each answer.
// first message
{ "thread_id": "new", "messages": [ { "role": "user", "content": "Hi, I'm Asha from Dehradun." } ] }
// every message after that - only the new one
{ "thread_id": "thr_rof6mpn5sg7wcpju5rx9rhyn", "messages": [ { "role": "user", "content": "What's my name?" } ] }
A thread belongs to the assistant that started it. Read it back with GET /v1/threads/{thread_id}/messages and delete it with DELETE /v1/threads/{thread_id}. When you use a thread, "Ask before acting" also keeps track of the conversation by itself.
For long tasks, add "async": true. Pearl replies at once with status 202 and a job ID (it starts with job_), then writes the answer in the background, usually within a minute. Collect it with GET /v1/jobs/{job_id} (its status goes from queued to running to done or failed), or let Pearl send it to your server with a webhook. Background answers can be combined with threads and JSON answers, but not with streaming.
A webhook is your own web address that Pearl sends a message to after each answer an assistant gives through the API, and when a background answer is ready or has failed. Set it up with the assistant's Webhook button in three steps: add your address (it must start with https://), copy the signing secret (it is shown only once), and send a test.
A POST with a JSON body. The type is assistant.reply, assistant.failed or test; the body also includes the assistant, the thread_id or job_id when there is one, and the full answer in response. These headers come with it:
| Header | What it is |
|---|---|
Pearl-Event | The event type, the same as type in the body. |
Pearl-Delivery | A unique ID for this event (it starts with evt_). If the same ID arrives twice, it is a retry: handle it once. |
Pearl-Signature | t=<time>,v1=<signature>, proof that the message really came from Pearl. |
The signature is an HMAC-SHA256 of the time, a dot, and the raw body, made with your signing secret. Work it out yourself and compare; also refuse messages older than 5 minutes.
<?php // PHP
$body = file_get_contents('php://input');
$sig = $_SERVER['HTTP_PEARL_SIGNATURE'] ?? '';
preg_match('/t=(\d+)/', $sig, $t);
preg_match('/v1=([a-f0-9]+)/', $sig, $v1);
$expected = hash_hmac('sha256', ($t[1] ?? '') . '.' . $body, getenv('PEARL_WEBHOOK_SECRET'));
if (!$t || !$v1 || !hash_equals($expected, $v1[1]) || abs(time() - (int) $t[1]) > 300) {
http_response_code(401);
exit;
}
$event = json_decode($body, true); // $event['type'], $event['response'] ...
http_response_code(200);
Answer with any 2xx status within 10 seconds. If your server does not answer or answers with an error, Pearl tries again after 1 minute, 5 minutes, 30 minutes, 2 hours and 6 hours, then stops. The Webhook panel lists recent events with their result and the next try. Lost the secret? Create a new one there; the old one stops working at once.
Rather than charging by obscure technical units, Pearl AI keeps things simple: your account has a credit balance, shown in plain terms, and every question you ask uses a small amount of it.
Every Pearl AI request on your account is recorded in one activity log, whether it came from the chat page, your API key or one of your assistants. Open it from Activity in the chat page's side panel, or from the usage card on your Pearl AI settings page.
If you only want a chat bubble that answers visitors, Pearl AI Chat does it with no code. Use the API when you want to build your own chat window, put answers into your own pages, or connect your website's own data.
Your API key (it starts with pai_) is like a password for your Pearl AI account. Anyone who sees it can spend your credits. So your website's server calls Pearl AI, never the visitor's browser:
fetch() from your page straight to products.pearlorganisation.in fails. That's deliberate: it keeps your key out of your pages. Always go through your server, as shown below.Your code sends only the visitor's messages. Instructions belong in the assistant: a message with the role system is refused.
Create the folder wp-content/plugins/my-pearl-assistant and, in it, a file my-pearl-assistant.php with the code below. Then add your key and assistant ID to wp-config.php (above the line that says That's all, stop editing), and activate the plugin under Plugins.
// in wp-config.php
define('PEARL_AI_KEY', 'pai_YOUR_API_KEY');
define('PEARL_AI_ASSISTANT', 'asst_YOUR_ASSISTANT_ID');
<?php
/*
Plugin Name: My Pearl AI Assistant
Description: Answers visitors with my Pearl AI assistant. The API key stays on the server.
Version: 1.0
*/
if (!defined('ABSPATH')) exit;
add_action('rest_api_init', function () {
register_rest_route('my-pearl/v1', '/chat', [
'methods' => 'POST',
'permission_callback' => '__return_true',
'callback' => 'my_pearl_chat',
]);
});
function my_pearl_chat(WP_REST_Request $req) {
// a simple limit per visitor: 20 messages a minute
$ip = isset($_SERVER['REMOTE_ADDR']) ? $_SERVER['REMOTE_ADDR'] : '';
$k = 'my_pearl_' . md5($ip);
$n = (int) get_transient($k);
if ($n >= 20) return new WP_Error('slow_down', 'Too many messages. Please wait a minute.', ['status' => 429]);
set_transient($k, $n + 1, 60);
$message = trim(wp_strip_all_tags((string) $req->get_param('message')));
if ($message === '' || strlen($message) > 2000) {
return new WP_Error('bad_message', 'Please type a message (up to 2,000 characters).', ['status' => 422]);
}
$thread = (string) $req->get_param('thread');
if (!preg_match('/^thr_[A-Za-z0-9]+$/', $thread)) $thread = 'new';
$res = wp_remote_post('https://products.pearlorganisation.in/api/v1/assistants/' . PEARL_AI_ASSISTANT . '/chat/completions', [
'timeout' => 60,
'headers' => ['Authorization' => 'Bearer ' . PEARL_AI_KEY, 'Content-Type' => 'application/json'],
'body' => wp_json_encode([
'messages' => [['role' => 'user', 'content' => $message]],
'thread_id' => $thread, // Pearl AI keeps the conversation, so send only the newest message
]),
]);
if (is_wp_error($res)) {
return new WP_Error('offline', 'The assistant could not be reached. Please try again.', ['status' => 502]);
}
$code = wp_remote_retrieve_response_code($res);
$data = json_decode(wp_remote_retrieve_body($res), true);
if ($code !== 200) {
error_log('Pearl AI error ' . $code . ': ' . wp_remote_retrieve_body($res)); // details stay in your server log
return new WP_Error('busy', 'The assistant is not available right now. Please try again later.', ['status' => 503]);
}
return [
'reply' => isset($data['choices'][0]['message']['content']) ? $data['choices'][0]['message']['content'] : '',
'thread' => isset($data['thread_id']) ? $data['thread_id'] : $thread,
];
}
// [my_pearl_chat] shows the chat box on any page or post
add_shortcode('my_pearl_chat', function () {
$url = esc_url(rest_url('my-pearl/v1/chat'));
return '<div class="mpa-box"><div class="mpa-log" aria-live="polite"></div>'
. '<form class="mpa-form"><input class="mpa-in" maxlength="2000" placeholder="Ask a question" aria-label="Your question">'
. '<button type="submit">Send</button></form></div>'
. '<script>(function(){var box=document.currentScript.previousElementSibling,log=box.querySelector(".mpa-log"),'
. 'form=box.querySelector(".mpa-form"),inp=box.querySelector(".mpa-in"),t=localStorage.getItem("mpa_thread")||"new";'
. 'function add(w,s){var p=document.createElement("p");p.textContent=w+": "+s;log.appendChild(p);log.scrollTop=log.scrollHeight;}'
. 'form.addEventListener("submit",function(e){e.preventDefault();var m=inp.value.trim();if(!m)return;add("You",m);inp.value="";'
. 'fetch("' . $url . '",{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({message:m,thread:t})})'
. '.then(function(r){return r.json();}).then(function(d){if(d.thread){t=d.thread;localStorage.setItem("mpa_thread",t);}'
. 'add("Assistant",d.reply||d.message||"Please try again.");}).catch(function(){add("Assistant","The assistant could not be reached. Please try again.");});});})();</script>';
});
Put [my_pearl_chat] on any page or post, then style .mpa-box to match your theme. The answer is shown as plain text, so nothing a visitor types can change your page.
Node.js with Express:
// npm install express (Node 18 or newer has fetch built in)
const express = require('express');
const app = express();
app.use(express.json({ limit: '20kb' }));
app.post('/api/chat', async (req, res) => {
const message = String(req.body.message || '').trim().slice(0, 2000);
if (!message) return res.status(422).json({ message: 'Please type a message.' });
const thread = /^thr_[A-Za-z0-9]+$/.test(req.body.thread || '') ? req.body.thread : 'new';
try {
const r = await fetch(`https://products.pearlorganisation.in/api/v1/assistants/${process.env.PEARL_AI_ASSISTANT}/chat/completions`, {
method: 'POST',
headers: { 'Authorization': `Bearer ${process.env.PEARL_AI_KEY}`, 'Content-Type': 'application/json' },
body: JSON.stringify({ messages: [{ role: 'user', content: message }], thread_id: thread }),
});
const d = await r.json();
if (!r.ok) { console.error('Pearl AI', r.status, d); return res.status(503).json({ message: 'Please try again later.' }); }
res.json({ reply: d.choices[0].message.content, thread: d.thread_id || thread });
} catch (e) {
res.status(502).json({ message: 'The assistant could not be reached.' });
}
});
app.listen(3000);
Python with Flask:
# pip install flask requests
import os, re, requests
from flask import Flask, request, jsonify
app = Flask(__name__)
@app.post('/api/chat')
def chat():
body = request.get_json(silent=True) or {}
message = str(body.get('message', '')).strip()[:2000]
if not message:
return jsonify(message='Please type a message.'), 422
thread = body.get('thread') if re.fullmatch(r'thr_[A-Za-z0-9]+', str(body.get('thread', ''))) else 'new'
r = requests.post(
f"https://products.pearlorganisation.in/api/v1/assistants/{os.environ['PEARL_AI_ASSISTANT']}/chat/completions",
headers={'Authorization': f"Bearer {os.environ['PEARL_AI_KEY']}"},
json={'messages': [{'role': 'user', 'content': message}], 'thread_id': thread},
timeout=60)
if r.status_code != 200:
app.logger.error('Pearl AI %s %s', r.status_code, r.text)
return jsonify(message='Please try again later.'), 503
d = r.json()
return jsonify(reply=d['choices'][0]['message']['content'], thread=d.get('thread_id', thread))
Your page then calls /api/chat on your own site, the same way as the WordPress box above.
wp-config.php, environment variables or a secrets store), never in HTML, JavaScript, a mobile app or a public code repository. If a key is ever exposed, delete it in Subscriptions → Pearl AI and create a new one."thread_id": "new" for the first message and the returned thr_… ID after that. Pearl AI keeps the last 40 messages of each conversation, so your code sends only the newest one.If something goes wrong with a request, the API replies with a status code explaining why. Here's what each one means and what to do about it:
| Code | What it means | What to do |
|---|---|---|
401 | Your API key is missing or no longer valid | Check the key is included correctly, and that it hasn't been regenerated or deleted since |
402 | The request can't be paid for. The code in the reply says why: credits_exhausted (no AI credits left), spend_cap_reached (your monthly spending cap is reached), plan_expired (your plan has ended), no_plan (no Pearl AI plan) or plan_upgrade_required | Top up credits, raise the cap or renew from Configure Pearl AI → Credits (or Billing), then try again. Your credits are kept while a plan is renewed |
403 | The key isn't allowed to do this: key_scope (the key is limited to other assistants) or ip_not_allowed (the request came from an IP address not on the key's list) | Change the key's access in Configure Pearl AI → API access, or use the right key |
422 | Something in the request isn't valid, for example an unknown model name or a missing assistant variable | Read the error message, which names the problem, fix the request and send it again. Common ones: invalid_request (for example a system message, more than 50 messages or 100,000 characters, an unknown model, or a thread_id that isn't new or thr_…) and missing_variables (the assistant needs a variable you didn't send) |
429 | Too many requests sent too quickly | Wait a moment and try again. Each API key can send up to 60 requests a minute |
404 | The assistant, thread or background answer was not found on your account | Check the ID you sent; a thread or job can only be read with the account that made it |
5xx | A temporary problem on Pearl AI's side | Try again shortly; if it keeps happening, contact support |
| You see | Why | What to do |
|---|---|---|
| You don't have Pearl AI yet | The signed-in account has no Pearl AI plan. | Press Purchase now →, or sign in with the account that bought it. |
| This Pearl AI subscription is not managed on this site. | The plan belongs to Pearl Client Workspace. | Open it from My Pearl Products → Configure Pearl AI. |
| Missing subscription reference. / Subscription not found. | The page was opened from an old or incomplete link. | Open it again from My Subscriptions or My Pearl Products. |
| Your plan allows N active assistant(s). Buy another assistant, or switch one off first. | Every assistant place is in use. | Switch one off (More → Switch off) or press Buy another assistant. |
| Please give the assistant a name. / Please write the instructions for this assistant. | A required box is empty. | Fill it in and save again. |
| Instructions can be up to 20,000 characters. | The instructions are too long. | Move facts into Knowledge, then save. |
| Variable "x" can use only letters, numbers and _ … | A {{name}} has a space or symbol. | Rename it, for example {{first_name}}. |
| We could not read any text from that file … | The file is scanned or a picture. | Use a text-based PDF, DOCX or TXT under 5 MB. |
| Remove one to add another. | Knowledge already has 5 documents. | Remove one, then upload. |
| Could not read this website … press Read again now. | The site needs a login, blocks reading, or the address is wrong. | Check the address opens in a private browser window without signing in, then press Read again now. |
| Does not match the format yet / The answer was not JSON. | The test answer didn't follow your answer format. | Keep Update the instructions too on, save, and test again. |
| Revoke a key to create another | You have 10 active keys. | Revoke one you no longer use. |
| That code did not work. | An old or mistyped security code. | Press Send a new code and use the newest one. |
| Your monthly limit has been reached, so new requests are paused until … | Your own spending limit was reached. | Raise or remove the limit on the Credits tab. |
| Payment confirmation failed — contact support with your payment ID. | You paid but the confirmation was interrupted. | Don't pay again. Send the payment ID to support. |
| Card payments are unavailable right now — try Razorpay instead. | The secure card form didn't load. | Refresh the page or use Razorpay. |
| View only. … | You are signed in as a team member without permission for this. | Ask the account owner. |
| One moment. Trying again in N s… | The server is busy. | Wait; it retries by itself. |
| This is busy right now. Please try again in a minute. | Still busy after the retries. | Wait a minute and press the button again. |
| Microphone access is blocked … | The browser blocked the microphone in Try it. | Allow the microphone for this site, or type instead. |
For errors your app receives from the API (401, 402, 403, 422, 429), see Errors & limits.
Pearl AI Chat is a chat agent — often called a widget, meaning a small self-contained piece added to a page — that you add to your website, app, CRM, ERP or any other place your customers use. Visitors see a chat bubble, usually in a corner of the screen, and can ask it questions about your business at any time of day.
Unlike a general AI chatbot, it only answers using information you give it — your own text, your policies, your product details, or documents you upload — so answers stay accurate to your actual business rather than guessing.
Everything about it is managed from one place: a settings screen called the configurator. Its tabs are Setup, Design, Knowledge, Tools / MCP, Connect, Domains & Embed, Conversations, Credits and Privacy & history. Each is covered below, starting with setting it up from start to finish.
www if your site uses both. The chat only appears on the addresses listed there.</body> tag near the bottom of the page.<script
src="https://clientworkspace.pearlorganisation.in/pearl-chat-widget.js"
data-key="YOUR_ACTIVATION_KEY"
async>
</script>
Using WordPress, Wix, Shopify, Squarespace, Webflow or a web app? See your website builder, step by step.
Once this is pasted in and saved, the chat bubble appears on your site automatically within a few seconds of a page loading — there's nothing else to install. From here on, everything you change in the configurator (the welcome message, what it knows, and so on) takes effect immediately, with no need to touch your website's code again.
Before you paste the code, open the Domains & Embed tab and add your website address. The chat only appears on the addresses listed there. Add both versions if your site uses them, for example example.com and www.example.com; each counts as a separate address.
Then copy your code from the same tab. It looks like this, with your own key in place of YOUR_ACTIVATION_KEY:
<script src="https://clientworkspace.pearlorganisation.in/pearl-chat-widget.js" data-key="YOUR_ACTIVATION_KEY" async></script>
The code goes once on every page, just before the closing </body> tag. Most website builders have a box for this. Find yours below.
Choose one of these ways. You don't need a plugin from Pearl, because the code works on any WordPress site.
footer.php, paste the code just above <?php wp_footer(); ?> or </body>, then update the file.If you use a caching or speed plugin (such as WP Rocket, LiteSpeed Cache or W3 Total Cache), clear its cache after saving. If the plugin delays or combines scripts, add pearl-chat-widget.js to its list of scripts to leave alone.
theme.liquid.</body> and save.Open Code Injection in your site settings (under Advanced or Website Tools, depending on your version), paste the code into the Footer box and save. Code Injection is available on Squarespace's Business plan and higher.
Open Site settings → Custom code, paste the code into Footer code, save, then publish the site. The code only goes live once you publish.
Paste the code into every page, just before </body>. If your pages share a footer file (for example footer.php or an include), paste it there once.
Load the code once for the whole app. In a plain React app, paste it into public/index.html just before </body>. In Next.js, add it to your root layout:
import Script from 'next/script';
// in app/layout.js, inside <body>
<Script
src="https://clientworkspace.pearlorganisation.in/pearl-chat-widget.js"
data-key="YOUR_ACTIVATION_KEY"
strategy="afterInteractive"
/>
Your activation key is meant to be public: it only works on the addresses listed in the Domains & Embed tab, so it's safe in your page code. This is different from a Pearl AI API key, which must always stay secret (see your own assistant on your website).
Work through these checks in order. Almost every case is one of them.
www.example.com and example.com are different addresses, and so is a staging or test site. With no addresses listed, the chat stays hidden everywhere.pearl-chat-widget.js. If it isn't there, the code wasn't saved, the page uses a different footer, or a cache is still showing the old page. Clear your site's cache and your browser's cache, then reload.data-key with the key in your configurator. For a second or later agent, use that agent's own key.https://clientworkspace.pearlorganisation.in for scripts and connections.Still stuck? Email support@pearlorganisation.com with your website address and a screenshot, and we'll check it for you.
Open it with Configure assistant on the Pearl AI Chat card in My Subscriptions (PEARL PRODUCTS site), or Configure Pearl AI Chat in My Pearl Products (Pearl Client Workspace). In Pearl Client Workspace, the header also shows your credits used, the reset date and Buy more credits →.
The tabs are grouped in the order you need them:
| Group | Tab | What it's for |
|---|---|---|
| Build | Setup | A ready template or your own name, brand and starter questions; version history. |
| Design | How the chat looks and behaves on your website, with a live test. | |
| Knowledge | Your agent's brain: business details, questions and answers, a document and your website. | |
| Tools / MCP | Actions it can take (bookings, order look-ups) and which ones ask the visitor first. | |
| Go live | Connect | How visitors reach your team: WhatsApp, phone, email, booking links and more. |
| Domains & Embed | Turn the agent on, list allowed websites, copy the embed code, set up the webhook. | |
| Manage | Conversations | Every chat, lead, rating and unanswered question; reply yourself. |
| Credits | Credits left, top-ups, spending limits and auto top-up. | |
| Privacy & history | Change history, how long data is kept, download or delete it. |
[not provided yet].On Setup, Design, Knowledge, Tools / MCP and Connect, the right side shows your real agent (on phones, press Preview). Replies use your knowledge and your credits. Save once before using it; until then it says This assistant has not been configured yet (Save your settings once before using the preview on the PEARL PRODUCTS site). The Test console under it lists every tool the agent used while you tested, what it sent and what came back; keys and passwords are hidden.
This tab sets how your agent introduces itself. From top to bottom:
#0751E0) and Logo (PNG, JPG, GIF or WEBP, up to 2 MB; square images work best). On the PEARL PRODUCTS site this card also has Widget enabled, the switch that turns the chat on or off wherever it is installed.The welcome message, action cards, colours and window are now set on the Design tab.
Ready templates are trained agents for every major industry, team and personal need. On Setup, press Browse templates (or a quick pick), or Full library to see them all on a separate page. The guided setup has five steps; each turns green when done.
Design changes how the chat looks and behaves on your website, never what it says. Every change shows at once in the Live test and in the page preview, but visitors only see it after you publish.
This is the most important tab, marked Your agent's brain. The agent answers only from what you add here; it never makes things up about your business.
https://www.yourbusiness.com).Only public pages of that one website are read: nothing behind a login, no files, and nothing your robots.txt blocks. For each question the agent looks at the 3 best-matching pages, which can use slightly more credits. Read again now refreshes at once; Remove website asks first and keeps your other knowledge.
Tools let your agent look things up or do things in your own systems during a chat, for example check an order or book a time. The tab guides you in three steps: Create a tool or add an MCP server, Connect it to your agent, and Try it in the live preview.
Under Start from a template, pick Look up an order, Check free slots, Book an appointment, Create a support ticket, Save a sales lead, Check product stock, Send to Zapier or Send to Make. Paste your own https link (keep the part in curly brackets; the agent fills it in), add a sign-in header if your system needs one, and press Add this tool.
check_order_status) and a Description that says what it returns and when to use it. The agent reads this to decide.https://), and the timeout.For MCP servers, press + Add MCP server, paste the server address (it usually ends in /mcp), add an access header if needed, press Connect & load tools, tick the tools your agents may use, and save. If the server needs you to sign in, press Sign in and approve access in the pop-up; your password stays with the provider.
For tools that change something (book, cancel, order, send), switch on Asks the visitor to confirm first. The agent then tells the visitor exactly what it will do and waits for a yes, which Pearl checks itself. Tools that only read run without asking. Recent calls shows the last 30 calls from live chats, the preview and tests.
Let visitors reach a real person right from the chat. Your agent also shares these whenever a visitor asks for a human; that is always on.
With several agents, Copy from Agent 1 copies the whole list (asks first; press Save to keep it). Each item shows how many clicks it got.
example.com (https:// and page paths are removed for you). If your site also opens as www.example.com, add that too, and add any test site. Up to 50.</body>, with Copy. Each agent has its own key, and the page always shows the code for the agent chosen in the agents bar.Messages under the website box tell you exactly what to fix, for example Add the ending of the website name, e.g. example.com or example.in. or This name has two dots in a row. A name saved before these checks existed is marked in amber; remove it and add the right one.
Every conversation visitors have with your agent is kept in the Conversations tab of the Pearl AI Chat configurator. If you run several agents, pick one in the agent bar at the top and the list switches to that agent's conversations.
The Unanswered marks are the quickest way to see what your content is missing. Add the answer on the Knowledge tab and the agent will use it from then on.
You can step into any conversation and reply as your team, straight from the Conversations tab.
Team replies never use credits. Only the account owner can pause the AI or reply.
At the top of the Conversations tab there are four parts:
A webhook sends each finished conversation to an address on your own server, with an optional summary and the details you choose to pull out, so it can go straight into your CRM, helpdesk or records. Each agent has its own webhook.
Open the Domains & Embed tab and find the Webhook block. If you run several agents, pick the agent in the agent bar first. Each step turns green when it's done and the next one opens:
https:// address on your server, for example https://yourdomain.com/pearl-webhook, then press Save.webhook.test) and shows your server's answer.Under What to include in each delivery, the agent can read the conversation and fill in details for you before it's sent.
customer_name: lowercase letters, numbers and underscores, starting with a letter, up to 40 characters (summary is reserved);YYYY-MM-DD) or One of a list (2 to 10 choices);Only what was actually said in the conversation is used. If something isn't mentioned, its value is null; nothing is guessed. The example delivery in the block updates to show your own fields.
Cost and limits. Pulling out a summary or fields uses the agent's normal credits, the same way a chat reply does, once per delivery. To keep it light, only the most recent part of a long conversation (about the last 8,000 characters) is read. If you haven't switched on a summary or added any fields, nothing extra runs and nothing extra is charged.
conversation.ended): after the visitor has been quiet for the time you choose (5 minutes to 12 hours; 15 minutes by default). Each conversation is sent once. If the visitor comes back and writes again, the whole updated conversation is sent again.conversation.manual): the Send to webhook button on any conversation in the Conversations tab sends it straight away, even while automatic sending is off. If the agent has no webhook yet, it offers a Set up webhook button that takes you to the block.A POST with a JSON body. Messages have the role visitor, agent (the AI) or team (a reply from your dashboard), and up to 1,000 messages are included. Inside conversation, visitor_ref is an anonymous ID that stays the same for the same visitor browser, and ai_paused is true when your team has taken over.
{
"event": "conversation.ended",
"delivery_id": "3f6c2a9e-8d1b-4c1e-9a52-7b0e1d4f9c21",
"sent_at": "2026-09-28T15:42:10+05:30",
"agent": { "number": 1, "name": "Support Assistant" },
"conversation": {
"id": 18342,
"started_at": "2026-09-28T15:31:02+05:30",
"last_message_at": "2026-09-28T15:36:48+05:30",
"message_count": 3,
"visitor_ref": "a1b2c3d4e5f60718",
"ai_paused": false
},
"messages": [
{ "role": "visitor", "text": "Hi, I'm Asha from Delhi. Do you ship to Canada? We'd order about 50 units.", "at": "2026-09-28T15:31:02+05:30" },
{ "role": "agent", "text": "Yes, we ship to Canada in 5 to 7 working days.", "at": "2026-09-28T15:31:05+05:30" },
{ "role": "team", "text": "Hi Asha, this is Priya from our team. Happy to help with your order.", "at": "2026-09-28T15:36:48+05:30" }
],
"extracted": {
"status": "ok",
"summary": "Asha from Delhi asked about shipping to Canada for about 50 units. The team followed up to help with the order.",
"fields": {
"customer_name": "Asha",
"city": "Delhi",
"quantity": 50,
"email": null
}
}
}
extracted is only included when you've switched on a summary or added fields. Its status tells you how it went:
ok: the summary and fields were filled in (anything not mentioned is null).empty: the conversation had no messages to read.no_credits: the agent's credits had run out, so nothing was pulled out.skipped, or any other value: extraction didn't run this time.Whatever the status, the rest of the delivery (the conversation and all its messages) is always complete.
X-Pearl-Event: the event name, the same as event in the body.X-Pearl-Delivery: a unique ID for this delivery.X-Pearl-Timestamp: when it was sent, in Unix seconds.X-Pearl-Signature: sha256= followed by an HMAC-SHA256 of timestamp + "." + raw body, made with your secret.Always check it before trusting a delivery. Use the raw body exactly as received, not re-encoded JSON, and reject timestamps older than 5 minutes. The same examples are in the Webhook block, under What we send and how to check it.
// Node.js (Express)
const crypto = require("crypto");
app.post("/pearl-webhook", express.raw({ type: "application/json" }), (req, res) => {
const ts = req.get("X-Pearl-Timestamp") || "";
const sig = req.get("X-Pearl-Signature") || "";
const expected = "sha256=" + crypto.createHmac("sha256", process.env.PEARL_WEBHOOK_SECRET)
.update(ts + "." + req.body).digest("hex");
const fresh = Math.abs(Date.now() / 1000 - Number(ts)) < 300;
if (!fresh || sig.length !== expected.length ||
!crypto.timingSafeEqual(Buffer.from(sig), Buffer.from(expected))) {
return res.status(401).end();
}
const event = JSON.parse(req.body);
// save event.conversation, event.messages and event.extracted
res.status(200).end();
});
<?php
// PHP
$body = file_get_contents('php://input');
$ts = $_SERVER['HTTP_X_PEARL_TIMESTAMP'] ?? '';
$sig = $_SERVER['HTTP_X_PEARL_SIGNATURE'] ?? '';
$expected = 'sha256=' . hash_hmac('sha256', $ts . '.' . $body, getenv('PEARL_WEBHOOK_SECRET'));
if (abs(time() - (int) $ts) > 300 || !hash_equals($expected, $sig)) {
http_response_code(401);
exit;
}
$event = json_decode($body, true);
// save $event['conversation'], $event['messages'] and $event['extracted']
http_response_code(200);
# Python (Flask)
import hmac, hashlib, os, time
from flask import Flask, request, abort
app = Flask(__name__)
@app.post("/pearl-webhook")
def pearl_webhook():
body = request.get_data()
ts = request.headers.get("X-Pearl-Timestamp", "")
sig = request.headers.get("X-Pearl-Signature", "")
expected = "sha256=" + hmac.new(os.environ["PEARL_WEBHOOK_SECRET"].encode(),
ts.encode() + b"." + body, hashlib.sha256).hexdigest()
if abs(time.time() - int(ts or 0)) > 300 or not hmac.compare_digest(expected, sig):
abort(401)
event = request.get_json()
# save event["conversation"], event["messages"] and event.get("extracted")
return "", 200
2xx status within 10 seconds. Do slow work after you've answered.conversation.id as the key and replace what you stored before.https addresses are accepted. Redirects aren't followed, and addresses on private or internal networks are refused.The agent automatically notices which language a visitor is writing in and replies in that same language, translating your knowledge on the fly. This means you only ever need to write your content once — typically in whichever language you're most comfortable with — and visitors elsewhere can still get an answer in their own language.
Your plan includes 1 agent. You can add more for $100 each, for example one for sales and one for support, or one per website, app or CRM.
agt_).Extra agents stay active until your plan's current end date. See Plans & pricing for how extras work.
The bar above the tabs shows one chip per agent and N of M agents. Click a chip to switch; every tab then shows and saves that agent.
Pearl AI Chat runs on its own credit balance, always shown in plain dollar terms. On the Credits tab you can:
If your credit runs out and isn't topped up, the chat on your website tells visitors it has reached its message limit for now, so you'll know to check this tab. Your team's own replies from Conversations never use credits.
The Privacy & history tab shows Change history (who changed what, marked You or PEARLian) and Data & privacy: how long each kind of data is kept, Download your data (JSON or CSV) and Delete now. Only the account owner can change, download or delete data.
| You see | Why | What to do |
|---|---|---|
| The chat doesn't appear on my website | Usually the website isn't listed, the code isn't on the page, or the agent is off. | Work through If the chat doesn't appear. |
| Fix the website name under Allowed Domains. | A website name has a typo, space or missing ending. | Type only the name, for example example.com, then save. |
| Business info is too long … | More than 50,000 characters of knowledge text. | Shorten it, or move detail into the document or your website pages. |
| We could not read any text from that document … | A scanned file or picture. | Upload a text-based PDF, DOCX or TXT under 5 MB. |
| That logo file could not be verified as a valid image. | The file isn't a real picture. | Use PNG, JPG, GIF or WEBP under 2 MB. |
| This assistant has not been configured yet. / Save your settings once before using the preview. | The agent has never been saved. | Press Save, then try the preview. |
| The agent could not reply. | Often no credits left, or a spending limit reached. | Check the Credits tab. |
| You are using all the agents in your plan … | Every agent place is used. | Press + Buy another agent, or remove one. |
| Off — over plan | Extra agents ended at renewal. | Buy another agent, or remove an agent you don't need. |
| This agent can no longer be restored … | More than 30 days have passed. | Create a new agent. |
| Choose the country, then enter the number without the country code. | The phone or WhatsApp number includes the code twice, or no country. | Pick the country and type the rest of the number. |
| Test failed / Could not connect. (Tools / MCP) | Your server refused, timed out or needs sign-in. | Check the address, method and headers; for MCP, press Sign in. See Recent calls. |
| Webhook failed: … / Test failed · … | Your server didn't answer with a success in time. | Check the https address and secret, then Send to webhook again. |
| Your monthly limit has been reached … | Your own spending limit. | Raise or remove it on Credits. |
| Payment confirmation failed — contact support with your payment ID. | You paid but the confirmation was interrupted. | Don't pay again; send the payment ID to support. |
| Your session expired. Please reload the page. | You were signed out. | Reload and sign in. |
| One moment. Trying again in N s… / This is busy right now … | The server is busy. | Wait; it retries by itself. If still busy, try again in a minute. |
| Visitors: "Allow the microphone for this website…" | The visitor's browser blocked the microphone. | Nothing on your side; they can allow it or type. |
Pearl Workstation Agent is a small application you install on your team's work computers. It shows how each computer is used during working hours: active time, idle time and breaks, which apps and websites are used, and regular screenshots. It works on Windows and Mac.
It is priced per computer: $200 a month (or $1,920 for 12 months) for each computer it's installed on. Ten computers need ten on your plan, from 1 up to 500.
You also get an end-of-day report and instant alerts for idle time, overrunning breaks and computers that stop reporting, plus optional suspicious-activity and data-leak alerts — by email to the people you choose, or in Slack or Microsoft Teams.
Open the dashboard with Manage workstations on My Subscriptions (PEARL PRODUCTS site) or Configure Pearl Workstation Agent on My Pearl Products (Pearl Client Workspace). The Setup tab has the downloads. Your activation key is already built into the installer, so there is nothing to type or copy.
PearlWorkstationAgent-Installer.zip.PearlWorkstationAgent-Installer-Mac.zip.After that, the agent starts with the computer and runs quietly in the background. Each computer counts once towards your plan; removing one frees its place straight away. To take the program off a computer, use uninstall.bat (Windows) or uninstall.sh (Mac) from the same zip. If the download says This subscription is not active., renew the plan first.
The Setup tab shows N computers connected · M online now. Press Set up another computer and follow four steps; each turns green with Done and the next one opens:
When it connects, you see … is connected and View activity logs →. If it hasn't connected after 3 minutes, the page says We couldn't detect your computer yet and lists four checks: the computer is on and online; the agent was installed with Run as administrator; a Workstation ID was picked and saved; and Test Connection says REACHABLE & ACCEPTED. Fix what's missing and press Retry. Your progress is remembered in that browser; Start over begins again.
In Pearl Client Workspace, the Setup tab also lists Registered Workstations (host, system, last user, agent version, last seen, Online or Offline) and has Add More Computers: choose 1, 3, 5 or 10 (or any number from 1 to 100), then Pay by Card or Pay with Razorpay. When it's done you see Computers added. You can now connect N computers.
On the Team tab, tell the system who works on which computer, so reports show names instead of computer IDs.
Removing a department keeps its people (without a department). Removing a person frees their Workstation ID. When someone joins, leaves or moves, update this tab and the next reports follow the change.
On Managers & team you can give people a read-only view, with no Pearl account needed. They sign in with a code sent to their email.
The Working hours tab tells the system what a normal day looks like. Alerts and reports only look at activity during these hours and days. They apply to your whole account.
Press Save working hours. How many idle minutes count as "idle" is set on Reports & alerts (the Idle alert's Alert after (minutes idle), 5 to 120).
The Timesheet tab shows each person's hours at the computer, day by day. Choose From, To and a department (or a quick pick such as This month), press Show, and click a person to open their days (Worked, On leave, Holiday, Day off, No activity). Download CSV saves a spreadsheet that opens in Excel or Google Sheets. Times are in your account's time zone.
On Holidays, add company holidays and each person's leave (one day or a range of up to 31 days). On those days there are no "computer not checking in" alerts, and the timesheet shows Holiday or On leave instead of No activity.
Before switching emails on, tell your team in writing that their work computers are monitored during working hours, and what is recorded. Then fill in Who sends these emails (your business name) and tick I confirm my team has been told their work computers are monitored.
For each report, switch it on and add up to 10 email addresses under Send to these email addresses. Some have extra settings: idle minutes (5 to 120), grace time after a break (5 to 60 minutes), a time to send, or up to 3 times a day. Where shown, you can also send to each department's manager, tell the person concerned, or add a weekly summary on the last working day. A report marked Available soon can't be switched on yet. Press Save reports & alerts.
| What it tells you | When it's sent | |
|---|---|---|
| End-of-day report | Each person's active time, idle time, breaks and the apps and websites they used most, with a team summary. | Once a day, after working hours |
| Idle alert | Someone has been idle for longer than your idle minutes during working hours. | As it happens |
| Break overrun alert | Someone hasn't come back from a break within the grace time. | As it happens |
| Coverage report | Which computers checked in during the day and which didn't. | At the times you choose |
| Suspicious activity alert | Optional. A remote-access tool, such as screen-sharing or remote-desktop software, is used on a computer. | As it happens |
| Data-leak alert | Optional. Files may be leaving the company, for example through file-sharing websites or apps. | As it happens |
| Computer stopped reporting | A computer has sent nothing for 20 minutes during working hours (breaks, pauses and leave are skipped). | As it happens |
Switch on AI work summary here to get a short, plain-language summary of each person's day by email, and a weekly one on the last working day. Pearl AI summaries and alert checks use your Pearl Workstation Agent credits. See Health, AI summaries and productivity.
Alerts can also go to team channels. Create an incoming webhook in Slack (Apps → Incoming Webhooks) or Teams (channel ··· → Workflows → "Post to a channel when a webhook request is received"), paste the address, tick the alerts, press Add channel, then Send test. A channel shows Working, Failed, Not tested or Paused. A report must be switched on above to be sent anywhere; alerts marked (off) are switched off there.
The Health tab checks every computer every minute while it's open: Computers, Online now, Offline, Need attention and Out of date. Everything looks good means every set-up computer is checking in. If computers need attention, each card lists the problem and what to do. Outside working hours, computers being off is expected and isn't a problem. Not set up means a computer has an ID but has never checked in.
The AI summaries tab shows each person's day or week in plain words, with active, idle, productive and distracting time and their top apps. Summaries are written from activity totals and app names only; screenshots, window titles and files are never sent to the AI. Press Write today's summaries now any time, or wait for the scheduled email.
On Productivity, rate apps and websites as productive, neutral or distracting; anything not rated counts as neutral. Not rated yet lists what takes the most time, rated with one click for the whole team, and Add suggested rules adds sensible ones. A department rule wins over a whole-team rule, and a website rule wins over the browser it's opened in.
The Activity logs tab goes from your whole team down to one computer's day in three steps: Department, Employee, Activity. Use the arrows, the date picker, Today and Refresh to change day.
Screenshots are private. They open only through links that work for your signed-in account and expire within minutes. If one won't load, close the viewer and press Refresh.
The Privacy tab (Privacy & history in Pearl Client Workspace) decides what is recorded and when. The rules apply on our servers, so they work for every computer at once, with nothing to change on the computers.
Press Save privacy settings; changes apply to new activity from that moment (Undo throws away unsaved changes). The same tab also has Change history and Data & privacy (how long data is kept, download, delete).
The Setup tab lists every computer on your plan: when it last checked in, its agent version and whether it is Online (checked in within the last 6 minutes) or Offline. The meter at the top shows Computers used n / N.
Adding computers: in Pearl Client Workspace use Add More Computers on the Setup tab; on the PEARL PRODUCTS site use Add time / computers on My Subscriptions. Each extra computer is a fixed $200, is ready straight away and stays on your plan: your renewal includes all your computers (see Plans & pricing).
In plain terms, Pearl Workstation Agent records how the computer is used during working hours, so you can see how work time is spent:
You decide what is collected: screenshots on or off or blurred, window titles hidden, websites by domain only, and pause times. Each employee gets a notice page that explains what is recorded and can confirm they have read it. Managers can be given a read-only team view, and employees can see their own work.
This information belongs to your account only. Screenshots are private and open only through short-lived links for your signed-in account.
To free a place on your plan, for example when someone leaves or a computer is retired, open the Setup tab and press Remove next to that computer. It asks first. The agent on that computer stops reporting, and its place is free straight away for another install. Removing it from the list doesn't uninstall the program: run uninstall.bat (Windows) or uninstall.sh (Mac) from the installer zip on that computer.
| You see | Why | What to do |
|---|---|---|
| We couldn't detect your computer yet | The agent isn't installed as administrator, no Workstation ID was saved, or the computer is offline. | Check the four points on screen, make Test Connection say REACHABLE & ACCEPTED, then press Retry. |
| Test Connection doesn't say REACHABLE & ACCEPTED | No internet, or security software or a firewall is blocking the agent. | Check the internet, allow the agent in your security software, or download a fresh installer and install again. |
| No Workstation ID free in the list | Every computer on your plan is in use. | Remove an old computer on Setup, or add more computers. |
| Offline / Need attention (Health) | The computer is off, asleep, offline, or the agent was closed or blocked. | Outside working hours this is normal. Otherwise turn it on and check the internet; the card says what else to do. |
| Out of date (Health) | An older agent version. | Download the installer again and install it on that computer. |
| Activity at the wrong hours, or alerts at odd times | Wrong time zone or working hours. | Working hours tab → set the time zone, day start and end, and working days → Save working hours. |
| Idle alerts during lunch | The break isn't added. | Add it under Breaks, or tick During breaks under Pause recording. |
| No screenshots, or they look blurred | Save screenshots is off, a pause time is on, or blur is set. | Check the Privacy tab. In the viewer, close it and press Refresh if a link expired. |
| Reports or alerts not arriving | The report is off, has no recipients, or the confirmation box isn't ticked. | Reports & alerts: switch it on, add emails, tick the confirmation, save, and check spam. |
| A Slack or Teams channel says Failed | The webhook address is wrong or was removed. | Make a new webhook, remove and add the channel again, then Send test. |
| Names show as IDs in reports | The person isn't linked to their Workstation ID. | Team tab → edit the person → choose their Workstation ID. |
| A manager gets no code | No access for that email, spam, or resending too fast. | Check Who has access, press Resend, and wait a minute between codes. A code lasts 10 minutes. |
| This subscription is not active. | The plan has ended or isn't paid. | Renew it on Billing or My Subscriptions. |
| Renewal overdue / Grace period / Payment due | The renewal isn't paid. | Pay or renew, and turn on Renew automatically with a saved card. |
| Your sign-in has ended. Please sign in again to see activity. | You were signed out. | Press Sign in again. |
PEARL MARKETPLACEPEARL MARKETPLACE is a business directory. Buyers find trusted businesses by service and location and ask them for quotes. Businesses get a public profile that shows their services, reviews and contact options, and receive enquiries from buyers.
It uses the same account as PEARL PRODUCTS and PRSXA™: one email and password for all of them (see PEARL MARKETPLACE sign-in).
Prices are shown in the business's own currency, with an approximate amount in your currency next to them. Badges on a profile mean:
Instead of contacting businesses one by one, describe what you need once and the most relevant businesses receive your request.
Each completed step turns green. When you're done, the matched businesses receive your request by email and WhatsApp and reply to you directly. A business can stop receiving quote requests at any time with the link in each request.
Many businesses are already listed from public information. If yours is, claim it, so you control what it shows:
Not listed yet? Choose Register as a partner. If your email or number already belongs to a listed profile, you're sent to claim it instead, so you never end up with two profiles.
After signing in you land on your dashboard, with these tabs:
Insights show how many times your profile appeared in search, how many people opened it, and your reviews and enquiries over time.
| Free claimed profile | Premium plan | |
|---|---|---|
| Price | Free | $100 for 6 months, or $150 a year |
| Services shown | 1 | Up to 10 |
| Enquiries and reviews | Locked | Included |
| Insights | Locked | Full insights |
Check your spam folder, and that WhatsApp works on the number you entered (with the right country code). Wait a minute and try again, or choose the other way to receive it.
Claim the profile and correct them from the Profile tab. If you can't claim it, email support@pearlorganisation.com with the profile link.
Yes. Buyers never pay to ask for quotes or to contact businesses.
PRSXA (said "Praxa") is Pearl Organisation's everyday AI assistant, powered by Pearl AI. You use it in your browser or as an app, with no API key and no setup.
PRSXA and Pearl Organisation are registered trademarks of Pearl Organisation.
Every product you turn on creates its own subscription under your account, each with its own status, its own renewal date, and — for Pearl AI Chat and Pearl Workstation Agent — its own activation key. You manage all of them in one place, your My Subscriptions page (My Pearl Products in Pearl Client Workspace). See Activating a product, step by step.
At the top of Billing, one saved card pays for the renewals and top-ups you turn on, for every product. The card is kept securely by Stripe; we never see or store the full number.
If a renewal shows Paused, the note says why. Fix the card, then turn it on again or renew by hand. Team members see View only. The account owner can change auto-pay.
| Status | Meaning |
|---|---|
| Active | Running normally. |
| Renews soon | 5 days or fewer left in this period. |
| Grace period / Payment due | The end date has passed and the renewal isn't paid. See Renewals & grace period. |
| Cancelled / Expired | The plan has ended. |
| Not subscribed | You don't have this product; press Subscribe to add it. |
On the PEARL PRODUCTS site, the Team page lets people help with your account. They sign in with their own email and password; you can change what they can do or remove them at any time.
Type their email, choose the role and press Send invite. Each person shows Active, Invite sent or Invite expired, with Resend, Cancel invite or Remove. If someone has invited you, open their account from Accounts you can open.
Each product is its own subscription. Pay month by month, or for 12 months at once and save 20%.
| Product | Monthly | 12 months | Included | Extras |
|---|---|---|---|---|
| Pearl AI | $50, plus usage credits | $480 | 1 assistant (2 on the 12-month plan) | $20 per extra assistant |
| Pearl AI Chat | $100, plus usage credits | $960 | 1 agent | $100 per extra agent |
| PEARL MARKETPLACE premium profile | — | $150 a year (or $100 for 6 months) | Up to 10 services, enquiries, reviews and insights | Boosts, bought separately (see PEARL MARKETPLACE plans) |
| Pearl Workstation Agent | $200 per computer, plus usage credits | $1,920 per computer | Windows and Mac | $200 per extra computer |
The plan cards and the full comparison table are on the Pearl Products home page.
Switching from monthly to the 12-month plan starts the 12 months when your current month ends, so you keep the days you've already paid for. You can also add more time to an active plan; it's added after your current end date.
Every purchase, top-up and extra is listed on your Billing page, with its invoice ready to download.
Prefer to speak to someone first? Our team can send you a secure payment link for any plan or extra. Once it's paid, your account is set up automatically and your login details are emailed to you.
Plans renew automatically with the card saved at checkout (cards in India pay in rupees and may need an e-mandate; everyone else pays in US dollars). We email you 5 days before each renewal. If you turn auto-renew off, you get the same email with a link to renew by hand, and an unpaid plan has a 3-day grace period before it ends.
If the renewal isn't paid by the end date, nothing is switched off straight away: the plan gets a 3-day grace period to sort out payment. If it's still unpaid after that, the plan is cancelled.
When a plan renews, extra assistants and agents bought for the previous period end, so add them again if you still need them. Extra computers stay: a Pearl Workstation Agent renewal includes all your computers.
There's no long-term lock-in. To stop a plan, turn off auto-renew for it on the Billing page; it then ends on its current end date. If Pearl AI is cancelled, your API key and assistants stop answering once the current period ends. If Pearl AI Chat is cancelled, your agents stop responding on your site once the current period ends; if Pearl Workstation Agent is cancelled, its computers stop reporting once the current period ends.
https addresses.Want to check whether Pearl AI, Pearl AI Chat, or Pearl Workstation Agent is running normally right now — or see details of a recent issue? The system status page shows the current state of each product, a day-by-day uptime history, and a log of any past incidents, including when each one started and when it was resolved.
Plain-language definitions for terms used throughout this guide.
asst_. It goes in the request address, never in place of your API key.thr_, so your code only needs to send the newest message.You may see this while Google is still reviewing the app for personal accounts. The sign-in page belongs to Google, and it lists exactly what Pearl AI asks for. If you continue, press Advanced, then Go to Pearl Organisation. If Google says access is blocked, that part is not open to personal accounts yet; Google Drive works for every account.
Pearl AI never guesses an address. If it can't find the person in your emails, it asks you. Type the address in the chat, or type it on the draft card before you press Approve.
Open the Webhook block in the Domains & Embed tab and send a test. Your server must be on a public https address and answer with a 2xx status within 10 seconds. Automatic sending covers conversations that end after you turn the webhook on, once the visitor has been quiet for the time you chose. A conversation marked failed can be resent with Send to webhook. See Webhooks.
Check that the computer is online and the agent is running, then pick the right date and open the hour (or press View screenshots). Screenshot links expire after a few minutes for privacy, so if the viewer has been open a while, close it and click the line again.
Check your spam folder, or that WhatsApp is working on your number. Wait a minute and resend (up to 3 times). If you're still stuck, contact support from the email address on your account.
Check two things: that the embed code is pasted in correctly, just before the closing </body> tag, and that the domain you're testing on is listed in the Domains & Embed tab — the widget deliberately won't load on a domain that hasn't been approved there. The full checklist is in If the chat doesn't appear.
This means the system doesn't recognize your API key. Double-check it's included exactly as shown in Authentication, and that the key hasn't been revoked or passed its end date, and that the request comes from an IP address the key allows.
Make sure that computer has an active internet connection, and that the agent hasn't been closed or blocked by antivirus or security software. If you need to reinstall it, use the same activation key — the same computer is not counted twice.
Copy the code from the Domains & Embed tab and paste it in your site's footer code area. There are steps for every common builder in your website builder, step by step. You don't need a separate plugin from Pearl.
Keep the key on your website's server and let the server call Pearl AI; never put the key in your page. Your own assistant on your website has ready-to-use code for WordPress, Node.js and Python. If a chat bubble is all you need, Pearl AI Chat does it with no code.
That's expected: the API only answers server-to-server requests, so your key never ends up in a web page. Send the question to your own server and call Pearl AI from there, as shown in your own assistant on your website.
Pearl AI accepts only user and assistant messages. Put your instructions in an assistant's settings (see Assistants & variables), or pass changing details as variables.
Revoke that key at once in Configure Pearl AI → API access and create a new one, then update it on your server. A revoked key stops working immediately. Limiting keys to chosen assistants and IP addresses keeps a leaked key from being used elsewhere.
Yes. All three run independently under the same account, and turning one on has no effect on the others.
No — anyone can create an account and use any of these three products, regardless of any other relationship with Pearl Organisation.
Yes. Your plan includes 1 agent, and you can add more for $100 each. Each agent has its own knowledge, look and embed code, so you can have one for sales and one for support, or one per website.
Extra assistants and agents last until the end date of the plan period they were bought for. When the plan renews it goes back to what it includes, so add them again if you still need them. Extra computers are different: they stay on your Pearl Workstation Agent plan and are included in the renewal.
You have probably sent the assistant ID (it starts with asst_) as the key. The assistant ID goes in the address; your API key (it starts with pai_) goes in the Authorization header.
Open Manage workstations (or Configure Pearl Workstation Agent in Pearl Client Workspace) → Reports & alerts. Check the email is switched on, the right recipients are listed and the confirmation box is ticked, then press Save reports & alerts. Also check your spam folder, and that each person is linked to a computer in the Team tab.
If anything in this guide doesn't match what you're seeing, or you run into a problem it doesn't cover, contact support@pearlorganisation.com. Include the email address on your account and a short description of what happened — a screenshot helps too, if you can attach one.