Use the browser with CLIO

CLIO has a browser of its own. When the agent opens a web page, the page appears as a tab in the side panel. You see what the agent does there, and you can use the page yourself: click, type, scroll, or enter an address. The agent and you share the same page.

The agent uses the browser for pages a plain download cannot read, such as pages that need a sign-in, pages built by scripts, forms, and web apps running on your computer. For a public page it can read directly, it fetches the page instead.

A conversation where CLIO counted the rows of a dataset table, with the same page open in the Browser tab of the side panel, showing back, forward, reload and the address — The page the agent read, open in the side panel. You can click and type in it, or enter another address.

Use the shared page

Links in an answer. Choose a web link and it opens in the side panel. To open it in your own browser instead, hold Shift (or Ctrl, Cmd or Alt) as you click.
Taking part. Anything you do on the page is yours: the agent is told you used the page and reads it again before it continues.
More than one window. The first window that shows a page controls it. Another window, or another device, shows the same page to watch only.
Pages a conversation opened. The Work summary lists the pages of the conversation and of its subagents.
Closing. Deleting a conversation closes its pages. A subagent’s pages close when its work ends. A page nobody uses or watches closes after 30 minutes.

Which sites the agent may load

The first time the agent loads a site in a conversation, CLIO asks you, with the site’s name. You can allow it once, for the conversation, or for the workspace. While you decide, the agent’s time limit does not run.

Pages you open, links you follow, and a page’s own redirects (such as a sign-in that passes through another site) load without a question, as in any browser. If a page reaches a site the agent may not use yet, CLIO asks before the agent reads or acts on it. A page is refused only when its address is of a kind the browser does not open, or when CLIO cannot check the site; the panel then says so.

The panel opens only web pages. Local files and browser-internal addresses are refused.

Sign-ins, passwords and files

Sign in yourself. The agent does not type into password fields. When a site needs a sign-in, it asks you to sign in in the panel, then continues.
Sign-ins are kept. CLIO’s browser keeps its cookies between conversations and across restarts, so you sign in to a site once.
Uploads follow the same file permissions as the rest of CLIO.
Running a script in the page asks you first.
Screenshots and downloads are saved with the conversation’s files. A screenshot records the page it came from.

Devices with their own certificate

A printer, a home server or a lab instrument on your network often uses a certificate the browser does not trust. CLIO never gets around such a warning on its own: the agent stops, tells you the site and the certificate’s fingerprint, and asks you to look.

The page stays on the browser’s warning in the side panel, with a note above it. Hover over the note’s information icon to see who the certificate was issued to and by, when it expires, and its fingerprint. If you know the device, choose Trust this certificate for This conversation or This workspace. The browser restarts, the page loads, and the agent can continue.

The trust is for that certificate. If the device later presents a different certificate, CLIO asks you again.
The browser’s own Proceed link is not the same. Clicking through the warning lets you see the page until the browser closes, but the agent still treats it as untrusted.
Removing the trust. The trusted certificate is listed in Choose what CLIO may do, where you can remove it.
Your running Chrome. When CLIO uses your own Chrome, proceed in Chrome if you trust the site. CLIO keeps no record of it.

When a site does not answer

Pages load from the computer the CLIO backend runs on, not from yours. When a site does not answer, the agent says which computer tried, and for an address on a local network (such as 192.168.1.20 or printer.local) that the address was looked up on that computer’s network. A CLIO backend on a cluster or a remote server cannot reach devices on your home or lab network.

Choose the browser CLIO runs

Choose the browser in Settings, under Browser. The change applies when you save it: the pages open now close, and the next page opens in the browser you chose. The browser runs on the computer the CLIO backend runs on.

Browser	What happens	In Settings
Bundled Chromium (default)	CLIO downloads its own Chromium the first time the browser is used and keeps it in your user cache. It does not change any browser you have installed.	Which browser: A separate browser. Separate browser: Bundled Chromium.
Your installed Chrome or Edge	CLIO runs the Chrome or Edge you have installed, on a profile of its own. You sign in there once.	Which browser: A separate browser. Separate browser: Installed Google Chrome, or Installed Microsoft Edge.
Your running Chrome or Edge	CLIO opens its pages as tabs in the Chrome or Edge you are using, with your own sign-ins.	Which browser: Your running Chrome or Edge. Running browser to connect to: `chrome` or `msedge`.

In every case the pages show in the side panel the same way.

CLIO does not copy cookies or saved passwords out of your own Chrome: Chrome keeps them encrypted with your operating system’s keys. To work with your own sign-ins, let CLIO use your running Chrome:

Use Chrome 144 or later, and open chrome://inspect/#remote-debugging.
Turn on Allow remote debugging for this browser instance.
In Settings > Browser, choose Your running Chrome or Edge, enter chrome under Running browser to connect to, and save. In config.yaml the same choice is:
```
browser:
  backend: cdp
  cdp_url: chrome
```
Chrome asks you to allow CLIO when it connects. Stopping CLIO closes the tabs it opened and leaves Chrome running.

Browser settings

backend, channel and cdp_url are in Settings > Browser. All of them can be set in config.yaml under browser:, and each has an environment variable: CLIO_BROWSER_ followed by the setting’s name in capitals, such as CLIO_BROWSER_CHANNEL. A change to config.yaml or the environment applies after a restart.

Setting	Default	What it does
`backend`	`managed`	`managed` runs a browser for CLIO; `cdp` connects to a browser you run, at `cdp_url`.
`channel`	`chromium`	The browser `managed` runs: `chromium` (CLIO’s own), `chrome` or `msedge` (your installed one).
`cdp_url`	unset	`chrome` or `msedge` for your running browser, or a DevTools address such as `http://127.0.0.1:9222`.
`profile`	`default`	The profile whose sign-ins every conversation shares.
`granted_hosts`	unset	Sites every conversation may load without asking, such as `*.example.org`.
`max_pages`	`12`	Pages open at once. Opening one more is refused.
`idle_minutes`	`30`	Minutes before a page nobody uses or watches closes, and a browser with no pages.
`action_timeout_s`	`30`	Seconds one browser action may take. Time waiting for your answer is not counted.
`frame_width`	`1280`	Width in pixels of the side panel’s picture of the page. Lower it on a slow connection.
`frame_quality`	`70`	Picture quality, 1 to 100.
`allow_unsandboxed`	`false`	Run Chromium without its sandbox when the sandbox cannot start. The panel shows when it is on.

Web apps on your computer

The agent can start a web app, open it in the panel, and work in it with you. The marimo plugin does this for marimo notebooks: the agent starts the notebook’s server, opens it in the panel, and runs code in the notebook’s own Python session, so you see each change as it happens. marimo runs on CLIO’s own Python, so it works without Python installed on your computer and changes none of your environments.

The marimo plugin is not part of any agent by default. Add it to a conversation from the Agent panel, as described in Agents and plugins.

Next steps

Choose what CLIO may do

Agents and plugins

Configuration