Computer

Kaer Computer gives the agent a real machine to work on. Describe what you need, and it boots a clean, sandboxed cloud computer, then drives it the way a person would — browsing the web, filling in forms, writing and running code, working with real files — and carries the task through to the end. You watch all of it happen live on the computer's own screen, and you can step in at any point to steer, answer a question, or take the controls yourself.

Starting a computer

Type your goal into the composer — it reads “Ask AI Agents…” until you start typing — and press Enter to launch. Shift + Enter adds a new line instead of sending. Leaving the box empty and launching anyway still creates a computer, just named “New computer” with no task to start on.

Two starter chips, Deep Research and Deploy Agents, launch straight into a common starting point. More — a dropdown, not an inline expansion — holds a handful of other quick prompts plus Create custom agent…, which opens the same builder covered in Custom agents below.

An engine picker sets how much the agent leans on speed versus depth for the session:

Kaer still makes the final call on each individual step — this sets your general preference, not a hard override. Your choice is remembered, and you can change it later in Settings.

Where your browser supports it, a microphone icon sits next to the composer; recognised speech is appended to whatever you've already typed.

The send button is the animated Kaer orb with a white arrow inside it, and its motion changes as you type and again once launch begins — the button itself is the first sign something is happening. While the computer boots you won't see raw status text: the orb turns into a short sequence of plain-English lines instead, ending as your workspace opens:

Claiming a fresh machine…
Briefing your agent on your goal
Warming up the desktop…
Opening your workspace — watch it live

The same friendly tone carries over once you're inside — the boot checklist on the computer's own screen is written in plain language too, not a raw log.

Home: Chat Mode and Overview

The home screen switches between two modes with a segmented toggle at the top: Chat Mode, for launching and browsing, and Overview, the health check for everything already running. A pill beside the toggle — AI Agents Running N — jumps straight to Overview whenever something is active.

Chat Mode

Chat Mode is the composer's home: a greeting and headline, a billing banner if your account needs attention (see Limits and billing below), the chips and composer covered above, and an event banner underneath. Below that sit three category cards — Autonomous AI Agents, Business Intelligence, and Workflow Automations — grouping the kind of work Computer is good at, and a strip of four templates you can start from directly, the same gallery covered in Templates below.

The event banner is where Kaer flags something worth a look — a limited-time usage boost, a new capability, that kind of thing. What it's promoting changes over time; at the time of writing, for instance, it's flagging a week of double usage to mark Computer Mode going live. Click it to open the details in a sheet without leaving the page.

Overview

Overview answers three questions: what's running, what finished, and where the pressure is.

• Stat cards — System Efficiency (the share of your computers running healthy, tagged Stable, Watch, or Degraded), Task Success Rate (completed tasks as a share of everything that's settled, with a trend chart), and Credits Remaining (what's left against your total, this week's spend, and the trend).
• Your operating profile — a written profile, traits, and insights the agent has picked up about how you work, built from your own usage history and generated on-device. It appears once you've run a handful of tasks, and the full written summary may need an active plan.
• Usage patterns — one chart you can switch between three views: Activity by day, Time of day, and Run lengths.
• AI Performance Overview — Average Execution Time, Success Rate, and Automation Coverage, over the last 24 hours, 7 days, or 30 days, each measured against the period before it.
• Activity Log — every task with its time, agent, status, and duration; select any running ones to cancel them together.

Custom agents

Agents in the sidebar opens Computer's custom agents area — standing delegates you build once and reuse. Each one is just a name and a directive written in plain words: a deployment monitor, a PR reviewer, a market watcher, a QA sweeper, a research digester, or something built from scratch.

Delegate work on any agent's card asks for today's task, then launches it on a fresh computer running that agent's standing instructions alongside whatever you just asked for. Delete an agent from the same card once it stops earning its keep.

You can also create one without leaving the composer — More → Create custom agent… opens the same builder.

Templates

Templates in the sidebar opens a dedicated “Start from a template” gallery. Picking one pre-fills the home composer with that template's prompt — it never launches on its own, so you can still read and edit before you commit. The four-template strip on the home screen is a shortcut into the same gallery.

Inside a computer

Once a computer is running, its own page gives you four ways to watch it work, switched from a stage bar at the top:

A LIVE pill shows while the computer is active. Following agent keeps the stage switching automatically to wherever the agent is working; switch a tab yourself and your view pins there instead, until you turn following back on. Take control hands you the live screen directly, so you can click and type on the machine yourself.

A conversation panel runs alongside the stage: your original goal at the top, the steps grouped underneath as they happen, and — when the agent hits something it can't resolve alone — a Needs you prompt with a reply box right there. If you've read The Operator, this is the same pause as Confirm on the blob: work has genuinely stopped and is waiting on you. You can steer mid-run or send an entirely new instruction at any point, not only when asked.

The header carries the computer's current status — Booting, Working, Needs you, Done, or Stopped — plus how long the agent has actually been working on it. Click the title to rename the computer; a name like “Q3 competitor pricing” is easier to find again later than “New computer”.

Four run controls sit with the header:

Sidebar and navigation

The sidebar is the map of everything Computer holds:

• Home — back to Chat Mode and Overview.
• Agents — your custom agents.
• Templates — the template gallery.
• Open Computers — every computer you currently have open, each with a live status dot; close one from its × or with a middle-click.
• History — the full, searchable record of every computer you've created, whatever its status, so you can find and reopen any of them.
• Help & Docs and Settings, a rolling usage meter, and your account chip (name and plan), grouped at the bottom.

Search (⌘K) jumps straight to a computer by name from anywhere. The notifications bell lists recent computers and their status. Share copies a link to exactly what you're looking at. + New Computer focuses the composer so you can start typing immediately. An Upgrade to PRO prompt shows up on free and starter plans.

Settings

Settings at the bottom of the sidebar opens the same account-wide settings covered in Settings. Two things there are specific to how Computer behaves.

Autonomy decides how much a prompt does on its own:

Whichever you pick, approvals for sensitive actions always pause for your confirmation — that's enforced server-side, not a setting you can turn off. It's the same guarantee described in The Operator.

Usage limits shows two windows on the same budget: a rolling 5 hours and a trailing 7 days. Both draw on the same credit balance as the rest of Kaer, so it's one number to watch, not a separate Computer-only pool — see Billing and Usage.

The rest is quick to skim:

• Engine preference — the Auto/Fast/Powerful choice from the composer, saved here as your standing default.
• Notifications — a desktop alert when a background run settles; if it isn't showing up, check your browser hasn't blocked permission.
• Appearance — Light, Dark, or System, plus a Reduce motion toggle.
• Usage & billing — credits remaining against your total, your plan tier, recent spend, an overage note if you've gone past it, fixed top-up amounts ($10, $25, $50 — larger ones earn bonus credits), your billing portal, and recent invoices.

Limits and billing

Computer follows the same rule as the rest of Kaer: nothing silently bills. A few messages you might see:

• Out of credits — a run pauses and says so, the same way it would anywhere else in Kaer; top up or wait for the reset to continue.
• Computer limit reached — if you're already at your plan's limit on open computers, Computer tells you before starting a new one rather than failing partway through.
• Low-credit warning — a banner appears while your balance is getting low, ahead of actually running out, so a top-up is a choice rather than a scramble.
• Rate-limited — if requests are coming in faster than your plan allows, Computer says so plainly and tells you when to try again.

See Billing and Usage for how the underlying credit pool works.

Coming back to a computer

Every computer has its own address — /computer/w/:id — so you can bookmark or share it directly, and open tabs persist per device, picking up where you left off. /computer/agents, /computer/templates, and /computer/history open straight into those areas, and individual settings sections deep-link the same way.

A worked run

Say you want a competitor pricing sweep. Type the goal into the composer and press Enter — Auto already covers most days, so there's nothing else to pick. Watch the orb's status lines while it boots, then follow along in Screen as the agent opens a browser and gets to work. Partway through, it hits a login it can't get past alone: the header switches to Needs you, and a message asks which account to sign in with. Reply in the conversation panel, and the run picks up where it left off.

Once it's done, rename the computer to something you'll recognise later, and if you need the same sweep again next quarter, it's one search away in History.

Good to know

• Is this the same Computer mentioned elsewhere in these docs? Yes — this page is the full walkthrough of the Computer surface introduced in Surfaces.
• How is a custom agent different from an Assistant or an Automation? A custom agent's whole job is launching a fresh computer with your standing instructions when you delegate work to it. A general reusable persona for chat belongs in Assistants; work that should fire on its own schedule belongs in Automations.
• Does an idle computer cost anything? A computer whose machine is still running can keep drawing on your compute allowance even between tasks — that's exactly what Stop sandbox is for: it frees the machine while keeping the computer and its history, so you can pick it up again later. See Billing and Usage.
• Where's the account-wide activity log, not just Computer's? Activity covers every surface; Computer's own Activity Log in Overview is scoped to computers only.
• Can several computers run at once? Yes, within your plan's limit — each one runs independently, and Open Computers in the sidebar lists all of them.