AI infrastructure that ships, not slideware.

Two tiers on one key — pearl-4 for production accuracy and reasoning, pearl-4-mini for high-volume, latency-sensitive calls. Both share the same request format, so switching models is a one-line change.

Chat completions, a dedicated endpoint for each assistant, and endpoints to list your assistants and models — all on stable, versioned paths (/v1/chat/completions, /v1/assistants/{id}/chat/completions, /v1/assistants, /v1/models).

Create a separate assistant for each job, each with its own instructions and knowledge: notes, exact questions and answers, up to 5 documents (PDF, Word or text) and pages from your website. Give it tools and MCP servers so it can look things up and take actions. Pass variables such as a customer's name or plan in the request and they are filled into that assistant's instructions. Every change is kept in a version history, and saved test questions can be re-run after a change. Monthly plans include 1 assistant, yearly plans 2, and you can add up to 50 more for $20 each until your plan's current end date.

Each key handles 60 requests a minute. Going over returns a clear error you can retry — never a silent failure.

Usage is billed in dollars from a prepaid credit balance, and every response shows the credits used and the credits left. Top up anywhere from $5 to $2,000 from your dashboard.

REST-first, with copy-paste examples including Node.js and Python, and a ready-made Postman collection. Any standard chat-completions client library works by changing the base URL. Structured JSON output, streaming, server-kept threads, background jobs and signed webhooks are all built in.

Create as many API keys as you need. Each key is shown once, can be limited to chosen assistants and IP addresses, can have an end date and shows its own usage. Set a monthly spending cap and a low-balance alert, or turn on auto top-up.

Let AI desktop apps, code editors and other MCP apps read your Pearl AI plan, read-only, with a key you control.

Every request is recorded in one activity log in your workspace — workspace chats, API calls and assistant calls alike. Searchable, exportable, and kept for as long as your account is active.

Beyond the API, a ready-to-use chat workspace with voice and hands-free mode, image and live-camera input, live web search with sources and suggested follow-ups — plus Pearl Canvas (live previews of websites, apps, designs and diagrams with every version kept, and one-click publishing at a public link), real slide decks you can present or download as PowerPoint, spreadsheets with working formulas you can edit and download as Excel, Code Studio, Data Studio for CSV and Excel files, PDF Studio, 26 guided apps, Agent mode, scheduled Automations, a Library with branches, and a command palette with slash commands. Think deeper reasons step by step and lets you watch its thinking; answer styles include one learned from your own writing; it runs its own Python and JavaScript to answer from real results; Improve prompt sharpens a request before you send it. Memory and Projects with their own files keep your context, Search past chats picks up earlier conversations, and Temporary chat leaves nothing behind. It reads documents up to 20 MB and any public link you paste, writes brochures and reports as A4 pages you can download as PDF, places real photos in the websites and designs it builds, and shows when a reply is ready in your browser tab. Settings shows live usage and when it resets, billing and auto-reload in one place, Capabilities you switch on or off, Reflect (how you use Pearl AI over a week, a month or three months) and Time and focus (break reminders and quiet hours). Connectors link Gmail, Google Drive, Google Calendar, Google Sheets, Google Analytics, Search Console and YouTube, so it can find your emails, files and numbers and draft emails, meetings and sheet changes that wait for your Approve. Personal Google accounts have Google Drive now and the other parts once Google finishes its review. It also writes Deep Research reports and runs big tasks in the background while you are away, sets reminders and to-dos with a daily brief, builds complete multi-page websites that check themselves at desktop and phone size, shows a live Working panel of every step, reads scanned PDFs and audio recordings, makes interactive charts, calculators, quizzes and flashcards, keeps every version of an answer, writes long reports, plans and e-books with every section written at the same time, looks through your earlier chats when you mention them, and opens every new chat with your day: today’s to-dos and the chats worth picking up, with ideas from your own work in For you. It turns long pastes into tidy attachments, takes screenshots, reads PowerPoint, e-books, saved emails and long PDFs in full with page numbers, understands YouTube links, keeps everything it made in Creations, lets you save your own commands, move chats into projects, edit, download and import memories, run automations once, every few hours or only when something changes, save answers to Google Docs and Sheets, double-check the latest answer, and turn any answer into a to-do. Select part of an answer to ask about it or rewrite only that part, choose the websites web search looks at first or never uses, type your next message while an answer is still being written, and open long answers and Deep Research reports full screen to read, copy, print or save as PDF. Connect your own apps and systems through custom connectors, choosing for each action whether it may run, asks you first or never runs, set a monthly spend limit, see this week’s usage by feature, switch charts, file creation and connector suggestions on or off, and choose code colours, a code font and an interface font (Standard, Serif or Italic). New on 16 October 2026: real photos with their authors in answers, up to 4 pictures in a message with Mark up, document scanning to PDF, places with a map, live share prices, currency, a watchlist and sports scores, GitHub links and website checks, sharing with chosen people or a password, Search everything, live screen sharing, Turn into, editing Word and Excel files, picture templates, Study mode, device notifications, shared projects, team commands, a connector directory, charts read inside PDFs and more voice languages. New on 19 October 2026: voice mode with your camera or screen, meeting notes, a live interpreter and questions about a video, editing part of a picture, Microsoft 365, Notion and Slack, Pearl AI on WhatsApp, Word, Excel and Outlook add-ins and a browser extension, automations that start when something happens, group chats and live documents, project memory, a study hub, trip plans and shopping comparisons, back ends that run in the preview, the app on your devices with passkeys and Teen mode, numbered sources and model choice, and publishing with versions, a forms inbox, stats, SEO and your own domain with secure https. New on 20 October 2026: Memory of up to 3,000 notes kept by topic, earlier chats found by what they were about, in-depth answers in one go, Code Studio that runs, checks and fixes its own code, the whole file whenever you ask for the full code, answers that understand what you mean and know exactly what your account includes, Meta Ads and Google Ads, Facebook Page and Instagram posting (opening for every account as Meta and Google finish their reviews), your own WhatsApp Business number with a live inbox (with a Pearl AI plan with AI credits), zoom in pictures and a voice mode that never goes quiet. With a Pearl AI plan with AI credits, voice mode also works as a Live call you can talk over, like a phone call, and voice works smoothly on Android and iPhone. New on 23 and 24 October 2026: replies sized to your message, Memory you can trust when you say remember that, emails written as text unless you ask to send them, code checked honestly before you rely on it, every step of an answer shown live in the chat Overview, Agent mode with your connected apps, long messages that fold, follow-ups that stay on your question, and Pearl AI inside VS Code, Xcode, Android Studio and Antigravity.

Live in every market Pearl Organisation serves today — the same key works whether your traffic originates in Mumbai, London, or São Paulo.

Upload PDFs, Word documents or text files (up to 5 MB each), write up to 20,000 characters about your business, add exact questions and answers, or sync 10, 25 or 50 pages of your website daily, weekly, monthly or by hand. Every file is safety-checked before the agent learns from it.

Run several agents from one plan — for example sales on your homepage and support in your help center. Each has its own training, look, website list and install code, all managed from one dashboard.

Set the reply tone (formal, friendly, concise), upload a logo and accent color, and the widget — including its chat button — matches your site rather than looking like a bolted-on plugin.

It answers from your own content and says so plainly when it does not know.

Auto-detects the visitor's language from their first message and replies in kind, without a manual language switcher.

Every visitor conversation, agent by agent, in your dashboard with search and date filters. "Unanswered" marks show exactly where your content has gaps, and a weekly Monday email lists them so you can add an answer to the knowledge in one click.

When a visitor asks for a person, the agent shows the ways to reach you that you choose — WhatsApp, call, email, website, booking link, location or social pages.

Connect up to 15 of your own tools and MCP servers per agent (with sign-in where needed) so it can check orders, bookings or stock, and ask the visitor before it acts.

Optional lead form and "Was this chat helpful?" rating, both off until you switch them on, with an email to you for every new lead.

Pause the AI on any conversation and reply yourself from the dashboard. The visitor sees your reply in the same chat within seconds, and one click hands it back to AI. Team replies never use credits.

Turn on a webhook and every finished conversation is sent to your own server as signed JSON once the visitor goes quiet (you choose 5 minutes to 12 hours) — with an optional summary and up to 8 details you choose, such as name, email or budget — ready for your CRM, helpdesk or records. Send any single conversation by hand, and check your setup first with a test delivery.

One script tag. No plugin marketplace, no build step — paste it before </body> on your website or inside your app, CRM or ERP, and the agent is live.

The widget loads only on the domains you list, so your install code can't be reused on someone else's site.

Preview any agent on a sample website before you install it — a 15-minute demo link, no code needed.

Trained content is understood and answered across major world languages, not just the language it was originally written in.

Active vs. idle time logged automatically, with application usage grouped into categories — no manual clock-in required from employees.

Group computers into departments and link each one to an employee's name and code. Computer IDs are issued automatically, so there is nothing to type during setup.

Connecting a new computer walks you through each step, turning green as it's done — and the dashboard spots the new computer on its own the moment it connects.

Open a department, then a computer, to see an hour-by-hour timeline of apps, idle time and power events, with a date picker and filters.

Every screenshot appears as its own line in the activity log, with the time and employee. Click the line to open it in a full-size viewer with next/previous, zoom and keyboard navigation.

Break windows and overtime thresholds are configurable per team, with automatic flags when someone runs over — the same engine used internally.

Managers see each department at a glance — computers online now, today's active and idle time, and who needs attention — without opening individual reports.

A daily end-of-day report plus instant idle, break-overrun and "computer stopped reporting" alerts, and optional suspicious-activity (remote-access tools) and data-leak (file-sharing sites) alerts — by email under your business name, or in Slack or Microsoft Teams — with optional Pearl AI daily and weekly summaries.

A single installer for Windows and Mac with no IT ticket required. Send employees the link or install it yourself — under two minutes per computer, from 1 to 500 computers on a plan. Add more computers from your dashboard any time.

Activity and screenshots are visible only to your signed-in account. Screenshot images are never stored at a public address — each one opens through a private link that expires within minutes. Turn screenshots off or blur them, hide window titles, record websites by domain only, set pause times, and give each employee a notice page they acknowledge. Managers get a read-only team view, timesheets, a holidays and leave calendar, and employees can see their own work.

$200 per computer a month, or 20% off on the 12-month plan. Extra computers are a fixed $200 each and stay active until your plan's current end date.

One agent build for Windows and macOS, so a mixed fleet doesn't mean two separate rollouts or two support playbooks.

Works the same for a team in one office or a workforce spread across time zones — every computer reports into the same dashboard regardless of location.