ego (lite) is just a browser, ego is your personal agent across devices.
Join waitlist
Download for Mac
English
EnglishFrançaisDeutsch简体中文繁體中文日本語한국어РусскийPortuguêsEspañolItalianoहिन्दी
Solutions
DocsEnterpriseChangelogBlog

How to use ego (lite) with your own Agent Harness

Connect your own Agent Harness to ego (lite). Check where commands run, grant host execution only when needed, load ego-browser, and complete a browser task.

llms.txt

Example footage: This guide uses OpenClaw recordings to demonstrate one Agent Harness connection flow. Your Harness may use different menus, commands, and permission names.

Your own Agent Harness can control ego (lite) as long as it can run ego-browser on the computer where ego (lite) is installed. First identify whether you are using a Desktop interface or a TUI so you know where to look for settings. Then determine where the Harness executes commands. The second question decides whether host-execution access is required.

Treat the video as one possible Desktop workflow, not as a universal Harness interface.

Before you start

Install ego (lite), finish onboarding, and make sure your Harness can run shell commands.

  1. Download the DMG and open it
  2. Follow the instructions and double-click the ego (lite) icon to install
  3. Open ego (lite) and finish onboarding
    1. Open your Harness in its Desktop interface or TUI
    1. Confirm that the Harness can run commands on the computer where ego (lite) is installed

ego (lite) installs the shared ego-browser Skill under ~/.agents/skills/. If your Harness supports Agent Skills, configure it to scan that directory or import the Skill using the method documented by the Harness.

After installing ego (lite), fully quit and relaunch the Agent app before continuing. This lets the app reload the newly installed ego-browser Skill and show its corresponding slash command in apps that support Skill slash commands. Closing only the current session or window may not reload the Skill.

Desktop and TUI determine where settings live

A Desktop app can run commands directly on your computer or inside a sandbox. A TUI can do either as well. Use the interface type to find the relevant settings, then inspect the command runtime. Whether Bypass permissions are required depends on whether the Agent can execute commands on the computer where ego (lite) is installed.

  • In a Desktop interface, look for permissions under the session, workspace, Agent, Tool, or security settings.
  • In a TUI, check the startup flags, configuration file, environment variables, or session status line.

The next step determines which permission path applies to your Harness.

Step 1: Check where your Harness executes commands

First ask the Agent to run this command from the same session that will perform the browser task:

command -v ego-browser

This command only confirms whether the current session can find ego-browser. It does not prove that commands run on the host computer. Use the Harness session information or official documentation to identify the runtime, then confirm it with the real connection test in Step 4.

Choose the next step based on the execution environment:

  1. Commands run directly on this computer without a sandbox: ego-browser can launch the locally installed ego (lite). There may be no Bypass permissions setting, and you do not need to create one. Continue to Step 3.
  2. The Agent runs inside a local sandbox: the Harness must provide an official host-execution, unsandboxed, or host-bridge capability that lets the Agent hand ego-browser to this computer for execution. Continue to Step 2.
  3. The Agent runs in a container, virtual machine, remote server, or cloud runtime: the Harness needs a host bridge that can execute ego-browser on this computer. Without that bridge, installing or loading the Skill inside the remote environment is not enough to control the local ego (lite) app.

If command -v ego-browser returns no path, first confirm that ego (lite) onboarding is complete. Then check whether the Harness is using a different PATH or a different execution environment from your normal terminal.

Step 2: Make sure the Agent can execute outside the sandbox

Before continuing, make sure the Agent can execute outside the sandbox and run ego-browser on the computer where ego (lite) is installed. If the Agent already runs directly on this computer without a sandbox, no additional permission is needed.

If the Harness asks for confirmation before each command, you can approve commands manually or enable its no-approval mode for a trusted session. No-approval mode only reduces confirmation prompts; it is not required to connect to ego (lite).

Desktop example: Approval policy

This recording shows OpenClaw's Exec Approvals settings. It only demonstrates how a Desktop Harness may manage command approvals. Turning off approval prompts does not prove that the Agent has host-execution access. If your Harness uses a sandbox, first follow its official documentation to enable host execution or a host bridge, then adjust the approval policy if needed.

TUI example: No-approval mode

This recording shows OpenClaw applying its own no-approval preset before opening the TUI. It changes OpenClaw's approval policy; it does not mean another Harness can now escape its sandbox. Do not copy the OpenClaw command into another Harness unless its documentation supports it. Look separately for the official host-execution method and approval setting provided by your Harness.

Host execution expands what the Agent can access on your computer. No-approval mode removes the confirmation step before each command. These controls carry different risks. Enable them only for a Harness, workspace, and task you trust. If the Harness supports a host-command allowlist or a narrower policy, prefer granting access only to the commands required for this workflow.

Step 3: Load the ego-browser Skill

Desktop example

The recording uses OpenClaw's Skill picker to load ego-browser. In your own Desktop Harness, open a local, shell-capable Agent session and load ego-browser from its Skill or Tool interface. The exact command shown in OpenClaw is product-specific.

TUI example

If your TUI supports Skill slash commands, enter /ego-browser, add a space, and continue with the task in the same message. If it does not provide slash commands, ask the Agent to locate and read the ego-browser Skill before beginning. A missing /ego-browser menu item does not by itself mean the Skill cannot be used.

You can use this compatibility check as the first prompt:

First determine whether your command runtime is running directly on this computer or inside a sandbox, container, or remote environment. If it uses a sandbox, confirm that you have an official permission or bridge that lets commands run on this computer. Do not treat disabled approval prompts as proof of host execution.

Then locate and read the ego-browser Skill. First check ~/.agents/skills/ego-browser/SKILL.md. If you cannot execute ego-browser on this computer, stop and explain what host-execution permission or bridge is missing.

Step 4: Run a simple connection test

Before opening a signed-in site, test the connection with a simple public page:

Use the ego-browser Skill to open https://example.com in ego (lite), report the page title, and leave the page open.

The host-execution path works when ego (lite) opens an Agent Space, the page loads, and the Harness returns the page title. If the page opens after you approve the command manually, only the approval policy remains; you can keep approving commands or use the Harness's documented no-approval setting for a trusted session. If the command never reaches this computer, return to Steps 1 and 2 and check the host-execution permission or bridge.

Step 5: Let your Agent complete a browser task

Once the connection test works, give the Harness a real browser task. The recording uses OpenClaw as the example controller, but the visible ego (lite) workflow is the same: the Agent loads the Skill, opens an Agent Space, visits the target page, and gathers the requested information.

Try the same task used throughout these guides:

Use the ego-browser Skill to open Anthropic's official X profile and collect the first five posts shown in the Posts feed.

If the Harness does not support Skill slash commands, this natural-language instruction is enough as long as the Agent can locate the Skill and execute ego-browser on the host computer.

Step 6: Check the completed result

When the task finishes, confirm that the Harness returned the requested information and that the source page remains open in ego (lite). The recording shows the result in OpenClaw; your own Harness may present the answer differently.

If the website asks for a login, CAPTCHA, or account verification, take over in the ego (lite) Agent Space and complete it yourself. Never put passwords or verification codes in the Agent prompt.

Troubleshooting

My Harness does not have Bypass permissions

That can be normal. If commands already run directly on your computer without a sandbox, there is no sandbox permission to bypass. Confirm that command -v ego-browser returns a path from the Agent session, then verify the connection in Step 4. If the Harness only asks for per-command approval, you can approve each command manually instead of looking for a Bypass permissions setting.

/ego-browser does not appear

Slash commands are a Harness interface feature, not a requirement of ego (lite). Ask the Agent to read ~/.agents/skills/ego-browser/SKILL.md, or configure the Harness to scan ~/.agents/skills/ using its supported Skill-import method.

ego-browser is not found

Finish ego (lite) onboarding, restart the Harness, and run command -v ego-browser again from the same Agent session. If the command works in your normal terminal but not in the Harness, compare their PATH and execution environments.

The Agent keeps asking for approval

This only means that the Harness is still applying its approval policy; it does not prove that the Agent is trapped in a sandbox. If ego (lite) opens after manual approval, keep approving commands or use the product's documented no-approval setting for a trusted session. If the command still cannot reach the host after approval, check the host-execution permission or bridge. Do not copy commands or flags from another product.

The Harness runs in Docker, SSH, or the cloud

The remote runtime cannot control a local ego (lite) installation unless the Harness provides a host-execution bridge. Run the Harness locally, configure its supported bridge, or install and run the compatible browser environment on the same machine as the Agent runtime.

The Skill loads but ego (lite) does not open

Confirm that the Agent is executing ego-browser on the computer where ego (lite) is installed. Loading the Skill gives the Agent instructions; it does not move command execution from a sandbox or remote machine onto your computer.

Go deeper