Skip to content

The agent home

On Pro, your agents keep working in the cloud while your devices are off. This is part of Cloud Sync. Your agents live in your agent home: your own small private database in the cloud, with a small program that sleeps for free and wakes in milliseconds. (The diagram calls it by its technical name, the cloud box.)

Inside Cloud Sync, your cloud box is the brain: data, memory, schedules, chats and the log. It borrows a browser, a sealed computer or an AI model only while a job needs one.Cloud Sync (Pro)Cloud box (the brain)your data and searchmemory, skills, agentsschedules and alarmschats and chat channelsthe log: every actionasleep for free, wakes at onceborrowed only while a job runsa web browsera sealed computer for codean AI modelsaved steps for long jobs
The cloud box is the brain, not a strong computer. It borrows a body for each job and gives it back.

What wakes an agent

  • A schedule: “every day at 8:00”.
  • A message: you write in the app, on Telegram or on WhatsApp. See Chat channels.
  • A webhook: another service sends data, for example a watch sync. What arrives is data, never an order to the AI.

Example: at 8:00 your Coach agent reads last night’s Sleep entry and this week’s Workouts, writes a plan for today, and sends it to your phone. Your laptop stays closed.

What your agents borrow

Every borrowed tool goes through one tool interface, so a new tool plugs in later with no rebuild. Now means the first version of agents in the cloud, which comes after the first release. Next comes after that, and later after that. The full list is on How agents in the cloud work (technical).

The job needs Your agent borrows When
a web search, or one web page web search and page reading Now
an AI model Kaizendex AI or your own key, both through OpenRouter; small jobs (search vectors, short summaries, sorting) go through OpenRouter too Now
a web browser a browser in the cloud: Research mode, then Acting mode (below) Next, then later
code, command-line tools, file conversions a sealed computer, one per task Later
hours of work saved steps that survive restarts Later

No AI model runs inside the agent home itself. It has too little memory and no graphics card.

When a service has a connector, such as email, your agent uses the connector instead of driving a browser.

The agent’s browser

Mode What it can do When
Research read any public page, follow links, take screenshots. No sign-ins, no forms, no downloads that run. Next
Acting sign in, fill forms and click buttons, only on sites you approved. Each site’s sign-in is sealed and kept for the next job. Every form it submits goes through the outbox. Later

By default, your saved sign-ins are in one list, each with Remove, and a sign-in unused for 90 days is dropped. Example: you tap Remove next to your running club’s site, and your agent is signed out of it.

Safe by design

  • What the agent reads is information, never an order. A page, a message or data from a webhook that tells the AI to do something gives it no new permission.
  • Every outside change goes through one gate, the outbox. An outside change is anything that changes something outside Kaizendex: sending, posting, buying, deleting or changing something on a site or service, a form the agent submits, and every call to an outside MCP tool, unless you marked that tool read only. No tool can skip the gate, and the outbox records every action. A read-only tool can still leak what it reads, so mark only tools you trust.
  • By default, what the agent starts on its own waits for your Approve tap on your phone. An automation you approved runs without a new tap each time; its approval card said so. Reading needs no approval in Standard. Strict adds checks on what the agent reads from outside. See AI safety settings.
  • By default, a changed outside tool is blocked. An MCP tool that changed since you approved it waits until you approve it again.
  • Some changes only you can make. The agent cannot change where a collection lives, its AI choices, sharing, keys or your AI safety settings. It can only suggest the change. The suggestion waits in the app until you approve it. See Changes only you can make.
  • Every change to your data is in the log and can be undone. Something that already left Kaizendex (a sent email, a post) cannot be unsent. That is why, by default, it waits in the outbox for your Approve.
  • One user, one home. Your agent home holds only your data.
  • The AI model is never sent your password. In Acting mode the agent uses a saved sign-in first: the site keeps you signed in. When a password must be typed, our cloud fills it in while the agent’s view of the page is paused. Other secrets, such as a key for a weather service, are added by our cloud on the way out.
  • You stay in charge: a read-only mode and an off switch, in Settings → Agent home, with the schedules, the approved sites, the approvals and the run history.

Example: Coach writes an email to your physio about this week’s Workouts. On Standard, the email waits in the outbox. It goes only when you tap Approve on your phone.

What your agent can read

Your data Your agent in the cloud
Cloud Sync (the default) can read it, if the collection’s AI setting allows it
Cloud Sync, end-to-end encrypted can read it only if you give the agent a key for that collection
This device only never: it never leaves its device

End-to-end encrypted means only your devices hold the key. Your agent is a registered device of your account, with its own keys, so these choices apply to it at once. When you give it a key, the app says so plainly: your agent runs on our cloud and can read that collection. See End-to-end encryption.

An end-to-end collection that your agent holds a key for shows a badge: Your agent can read this, with Remove. Example: you give Coach a key to your Diary, so it can spot bad nights. The Diary now shows the badge. Tap Remove to take the key back. Giving an agent a key is a change only you can make.

What it costs

Your agent home sleeps most of the time, so it costs little. A small monthly amount of agent work is included in Pro. Beyond it, browser and computer time comes out of your credits at cost. The app shows the cost before a long job starts. See AI credits.

  • A daily spending limit. By default, agents stop for the day when they reach their spending limit for that day.
  • No surprise charges. When your credits run out, your agents stop and the app offers a top-up. It never charges you extra by itself.