Install and use the Companion extension

Install and pair

Pairing mints credentials of the extension's own, so it stays signed in on its own from then on — you don't need a Cheqq tab open for agents to reach your browser, and closing the app for the day doesn't sign the extension out. Unpairing (Remove) revokes those credentials.

Each browser install is a separate device — rename it, Make primary, or Remove it from that same settings page. A name you give a computer stays put. Agents use your primary online device, and everything they do happens in the Cheqq Agent tab group. Use Check on a device to make it prove it's reachable right now — the same probe agents use before trusting a browser — so "is my browser connected?" is answerable without starting a task to find out.

Use it in chat

Open the extension from your browser's toolbar to reveal the side panel. From there you can capture the current page — its readable content, your text selection, or a screenshot — and hand it to a chat or an Agent-mode task.

What the agent can do in your browser

In chat and in a workflow's AI steps, the agent can browse the way you would — signed in, on real sites — and it checks its own work as it goes:

• It reads the whole page. Long emails and articles come through in full, a page too long to take in at once is read in parts, and frames the site embeds from itself are readable and clickable too.
• It sees its screenshots. In chat, when the model answering can read images, a screenshot the agent takes — of the view, one element, or the whole page — is something it looks at, not just a file it saved.
• It knows when an action did nothing. After a click or a key press it can tell whether the page changed, and tries another way instead of assuming it worked.
• It knows its way around Gmail. It picks the right account when you have several, searches with Gmail's own operators, opens a thread directly, and finds who is still waiting on a reply.

A long browser task keeps the same agent tab for as long as it runs.

Settings → Advanced → Browser Agent Runtime sets how long a workflow's browser step may run and how many steps it may take. An individual AI step can override both.

Use it in a browser-triggered workflow

A workflow's Browser trigger runs from the side panel against your current tab. Configure site filters (which pages it applies to) and what gets captured — page content, selection, or screenshot — when you run it.

The panel lists your browser-triggered workflows by name with an icon derived from what each one does, and shows a live run list underneath: every step appears under its real name with a status ring and duration, and a running workflow can be paused, resumed, or cancelled from the row.

Hand a file to a page

An agent driving your browser can upload a file into a page's file picker — a file from your workspace files, or an attachment on a record. It works on sites where a page would normally be blocked from reading your storage, and on pages the agent is working in the background.

This is what lets an automation finish the sentence "…and attach the image": Content Studio's poster uses it to attach media to a post, and refuses to publish a stripped version if the approved media doesn't attach.

Uploads are covered by the same consent as everything else the Companion does on your behalf, and a workflow's Browser trigger can list exactly which actions it's allowed to take.

Runs that wait for your machine

Unattended browser work needs a machine that's awake. When a scheduled workflow fires and no device answers, the run waits rather than failing — it parks, and starts by itself when your machine comes back, usually within a minute of the extension checking in again.

• Repeat firings coalesce. Six missed two-hourly sweeps produce one sweep when you return, not six — replaying old windows is worthless for exactly the work that needed to be early.
• Three skips in a row raise a notification. A skipped run leaves no trace in run history, because it never executed. This is the only thing that will tell you a schedule has stopped coinciding with your machine being awake.
• A run already in flight when the lid closes stops at once rather than retrying against a browser that isn't there, and picks up where it stopped by itself, a minute or so after that computer is back.
• A step that waited too long for a busy browser (another run or a chat had it) tries again by itself a few minutes later and carries on where it stopped.
• A stopped run carries on by itself within 12 hours, inside your active hours. After that it waits for you: resume or stop it from its run page, where it reads Waiting for your browser.

The side panel says what a parked run is waiting for: Waiting for your computer, Waiting for your computer to come back, or Waiting for your active hours, with the time it will run and how many missed runs it folds together.

Set the hours automations may run

Presence decides whether unattended work can run. Settings → Advanced → Active hours decides whether it should. Set a window and the days it applies to, and browser automations wait outside it — so opening your laptop at 2am to check one thing doesn't hand the Companion a queue of sweeps that takes over the session.

The window is evaluated in your account's timezone, not the browser's, and the preview names that timezone and says whether the window is open right now. Overnight windows (22:00 to 06:00) are described as such rather than as a range that reads backwards. It's off unless you switch it on.

Stay in control while the agent drives

When an Agent-mode task needs to control the browser directly — navigating, clicking, typing — it works inside a dedicated tab group, not your regular tabs. Pressing a button or link whose label is a command — Send, Delete, Pay, Place order, Submit — pauses for your approval in the side panel. Reading, scrolling, and opening an email or a result don't. If you don't approve, the agent tells you what it wanted to click and why rather than trying again. You can take manual control of the agent's tab at any point to pause it yourself.

You can see where it is and what it's doing. The agent's tab glows around its edge, and a cursor glides to whatever the agent clicks, hovers over or types into, with a ripple on each click. The Cheqq Agent tab group shows ⏳ while the agent is working and ✅ when it has finished. If your system is set to reduce motion, the glow stays steady and the cursor jumps instead of gliding.

The agent never uses Cheqq itself through your browser. Its tab refuses Cheqq's own pages, whether it types the address or follows a link, redirect, or form there, and returns to the page it was on. What the agent does in Cheqq, it does with its own tools, under your permissions and approvals. Apps you've published on their own domain are ordinary sites and stay reachable.

If a task reports it couldn't reach your browser, the message says which case you're in. Nothing paired means install the extension and sign in. Paired but asleep names the machine and asks you to wake it. If the side panel shows you as signed out entirely, sign in to the Cheqq web app once in the same browser to re-pair.