How CloudyBot works

CloudyBot is built for execution, not just chat. It opens a real browser, edits your files, runs on a schedule, chains AI workers that hand off to each other, and delivers results to WhatsApp or the dashboard. This page explains how each part works in plain English.

The core idea

Most AI tools require you to be present. You open them, type a prompt, wait for a response, close them. Nothing happens when you are not there. The work requires your attention at every step.

CloudyBot is designed around the opposite model. You describe what needs to happen. The AI plans the steps, executes them using real tools — a browser, your files, web search, external services — and delivers the result to wherever you are. You can be in a meeting, on your phone, or asleep. The work happens anyway.

That shift — from "AI that responds" to "AI that executes" — is what this page explains.

The cloud browser

The most distinctive thing about CloudyBot is that your AI has a real browser. Not a text scraper. Not a simplified HTTP request. A real browser running in the cloud that navigates actual websites the same way a human would — logging in, clicking buttons, filling forms, scrolling through dynamic content, extracting structured information from pages that would be invisible to a basic scraper.

When you ask CloudyBot to research five competitors, it actually opens their websites. When you ask it to find a LinkedIn profile, it actually navigates LinkedIn. When a scheduled Specialist monitors a competitor's pricing page every morning, it actually loads that page and reads it — comparing against what it saw yesterday to tell you what changed.

The browser is paired with a residential IP so it is not blocked by sites that detect automated traffic. Each session is isolated — your session's cookies, login state, and data are completely separate from any other user's session. Sessions are ephemeral by default: when the session ends, all state is cleared unless you have explicitly saved it to your workspace.

You can watch the browser work live in your dashboard. The live view streams in real time — you see every page it visits, every click it makes, every form it fills. If something looks wrong or it hits a captcha it cannot handle alone, you can take over the browser directly and hand control back to the AI when you are done.

Three-layer memory

Context window limits are the reason most AI tools seem to forget things. Every model has a maximum amount of text it can hold in active memory at once. Long conversations, large files, and complex tasks push against this limit constantly.

CloudyBot uses a three-layer memory system to solve this without requiring you to manage it:

Layer 1 — Active context

The current conversation, the last several messages, and any files or tool outputs from the current session. This is what the AI is actively working with right now — full fidelity, immediate recall.

Layer 2 — Rolling summary

As a conversation grows, older messages are progressively compressed into a rolling summary. The AI maintains the key facts and context from earlier in the conversation without keeping every message verbatim. This keeps token costs manageable while preserving meaningful context across long sessions.

Layer 3 — Pinned memory

Facts you explicitly tell the AI to remember permanently — your name, your business, your preferences, your ongoing projects. These are injected into every interaction so the AI always knows the context that matters most, regardless of how long ago you told it.

Scheduled Specialists also have their own memory of previous runs. When Scout checks a competitor's website on Wednesday, it compares against what it found on Tuesday and tells you the delta — what actually changed — rather than summarising everything from scratch every time.

Scheduled AI workers

Beyond interactive conversations, CloudyBot runs recurring autonomous workers (sometimes called Specialists in the product). These are AI workers you deploy once and they run automatically on a schedule you set — hourly, daily, weekly, or any custom interval — whether your device is on or not.

Each worker has a defined role, scheduled tasks, access to the tools it needs (browser, web search, files, integrations), and a delivery destination for its results. You describe the role in plain English. The Workflow Architect builds it and deploys it.

Pre-built Specialists you can activate immediately include:

Scout — web intelligence and competitor monitoring. Checks defined websites on a schedule, compares against previous findings, and delivers only meaningful changes.

Watchdog — site and page monitoring. Detects meaningful changes against a saved baseline and alerts you when something actually changes — not on every minor variation.

Analyst — data analysis and weekly reporting. Processes data from your workspace, produces trend analysis, and delivers structured reports on the schedule you set.

Postmaster — inbox triage and digest delivery. Reads your email before 8am, flags what is urgent, drafts replies for the threads that need a response, and delivers a prioritised summary.

Power Analyst — deep research with code execution. Runs Python scripts on your data, processes complex files, and builds reusable knowledge bases from repeated research runs.

Beyond pre-built Specialists, you can define custom roles with any duties your business needs. Any repeatable job that happens the same way every day or week is a candidate for a Specialist.

The Workflow Architect

Building a team of Specialists requires decisions: which roles, what duties, what schedule, where to deliver results, how specialists hand work to each other. The Workflow Architect handles all of this.

It is a specialist whose job is to understand your business, design the right team of AI workers, and deploy them — all through a conversation in plain English. You describe what needs to happen every week. It proposes a team with named roles, defined duties, and cron schedules. You review each specialist, adjust anything that needs changing, and deploy. Your first specialist can be running within fifteen minutes of starting the conversation.

Multi-agent pipelines

Specialists can hand work to each other automatically. A research specialist finishes its job and triggers an analysis specialist, which triggers a reporting specialist, which posts to your Slack channel. Each specialist picks up where the last one left off through shared workspace files.

The whole chain runs automatically — no one needs to initiate each step. The pipeline runs on a schedule you set once, delivers results to where your team works, and builds context over time as each run adds to the shared knowledge base.

File workspace

Every CloudyBot account has a persistent file workspace in the cloud. Files you upload — documents, spreadsheets, CSVs, PDFs — stay in your workspace across sessions and accumulate into a knowledge base over time. Research saved by Scout last month is available to Analyst this month when it produces its report.

The AI edits files surgically — changing only the lines that need changing rather than rewriting entire documents. This matters for your monthly task budget: editing three sentences in a 200-page document should not cost the same as summarising the whole thing. With surgical editing, it does not.

Storage scales with your plan: 50 MB (Free) through 5 GB (Max). Files remain private to your account and are never used for model training.

WhatsApp and mobile delivery

CloudyBot reaches you rather than waiting for you to check in. On paid plans, Specialists deliver results directly to WhatsApp — the same AI, the same threads, the same workspace, accessible from your phone like a contact in your address book. Proactive sends (results delivered to you without you asking) scale from 20/month on Growth to 500/month on Max.

The dashboard installs as a PWA on iOS, Android, and Windows — full workspace, same files, same specialists, same threads as the web version. Push notifications fire when scheduled jobs finish so you know when work is ready without checking.

Integrations

Specialists can deliver results to and read data from the services your team already uses. Native integrations include Slack (deliver to channels), Notion (append to pages and databases), Google Sheets (write structured data), GitHub (read repos, commit changes), Stripe (payment event triggers), and Zapier and Make for connecting to anything else.

These are native connections — the specialist calls the service directly rather than using the browser to navigate its interface. That means faster, more reliable delivery and less browser session usage for integration tasks.

Hard billing caps

Every CloudyBot plan has a hard monthly cap on AI Tasks and browser sessions. When you hit the cap, the service pauses. It does not charge you more. It does not silently keep metering. It pauses, shows you a clear message, and gives you the option to upgrade, buy a top-up pack, or wait for the next billing cycle.

AI Tasks are the billing unit — a whole number you can reason about. One standard message on a standard model is roughly one task. Premium models use multipliers shown in the model picker before you commit: standard tier is 1×, mid-tier (like GPT-4o) is 3×, Claude Sonnet is 6×, top-tier models are 15×. The same task costs more on a stronger model — and you always see that before you send the message.

Plans run from Free (30 tasks, no card) through Growth ($19), Growth ($19), Pro ($39), and Max ($149, 7,000 tasks). Your maximum monthly spend is always knowable before the month starts.

Privacy and data

CloudyBot does not train AI models on your data. Your conversations, files, and browser sessions are private to your account. We use your data only to provide the service — not to train anything or share with third parties beyond the subprocessors listed on our subprocessors page.

Each browser session is isolated — your authentication state and session data are completely separate from any other user's session. Sessions are ephemeral by default: nothing persists between sessions unless you explicitly save it to your workspace.

Frequently asked questions