How to Enable Computer Use for Warp Cloud Agents

If you send Warp cloud agents after UI changes, Computer Use is how they see the screen, click through the interface, and check whether the result matches. It is also easy to assume the feature is on when it is not. Cloud Agents you start from the Warp app ignore the server default. Claude Code, Gemini, and Codex runs stay off unless you enable Computer Use yourself. Use the official enablement paths below so a UI-test run actually gets a sandbox desktop.

Cloud sandbox only

Computer Use lets agents interact with desktop environments inside a secure, isolated sandbox. The agent can see what is on screen, click and drag, type text, use keyboard shortcuts, and perform other GUI interactions.

With Computer Use, agents can:

Take screenshots and analyze the current display
Record sessions as annotated video
Click buttons, fill forms, and navigate interfaces
Type text and use keyboard shortcuts
Automate UI testing end-to-end without manual intervention
Work with browser-based interfaces and the web

Warp’s headline use case is testing UI changes with a self-contained feedback loop: the agent verifies that your code changes produce the expected visual and behavioral results.

Computer Use is only available in Warp’s sandboxed cloud environments, not in local interactive terminal sessions. The sandbox is fully isolated. It does not have access to your local machine, credentials, or sensitive data outside the sandbox. Recordings stay inside the sandbox too—they never include your local desktop or credentials.

Browser-based tasks need no extra setup. Every Computer Use environment includes a bundled Chromium browser, which agents can drive visually or with scripted Playwright commands. Any other desktop application you want the agent to control must be included in your cloud environment configuration.

Computer Use currently supports Anthropic Claude models. Warp uses an auto model selector to choose the best-suited model for each Computer Use task.

Why the default depends on how you start the run

On Warp’s built-in harness, Computer Use is enabled by default. When a run’s configuration does not specify a Computer Use setting, the agent runs with Computer Use available.

Runs on third-party harnesses—Claude Code, Gemini, and Codex—default to Computer Use disabled. Those harnesses do not integrate with Warp’s Computer Use tooling. If you launch one of those cloud runs and never set Computer Use, the agent does not get Warp’s screenshots, clicks, typing, or bundled Chromium.

Cloud Agents started from the Warp app are a third case. They do not use the server default. They always follow the app’s Computer use in Cloud Agents setting (cloud_agent_computer_use_enabled), which is off by default.

That is the trap for teams doing UI tests from the desktop app: the built-in harness default is on, but app-started runs stay off until you change Settings.

Turn it on in the Warp app

To control Computer Use for Cloud Agents started from the Warp app, go to:

Settings → Agents → Warp Agent → Experimental → Computer use in Cloud Agents

The control lives under Experimental. Leave it off if those app-started runs should not drive a GUI. Turn it on when they should.

Control it per run with the Oz CLI

When you run cloud agents with the Oz CLI, set Computer Use per run with flags:

oz agent run-cloud --computer-use --prompt "<task>"
oz agent run-cloud --no-computer-use --prompt "<task>"

--computer-use makes GUI tooling available for that run. --no-computer-use turns it off. Replace <task> with the prompt you want the agent to execute.

Set `config.computer_use_enabled` on the API

When you create a cloud agent run with the Oz API and SDK, the optional config.computer_use_enabled field controls Computer Use. When the field is omitted, it defaults to true for runs on Warp’s built-in harness and false for runs on third-party harnesses.

The official example disables Computer Use for a run:

{
 "prompt": "Build a button component that matches this design, then test it in the browser",
 "config": {
 "computer_use_enabled": false,
 "environment_id": "optional-environment-id"
 }
}

If you are on Claude Code, Gemini, or Codex and you need the sandbox GUI, set this field explicitly. Omitting it keeps Computer Use off on those harnesses.

Configure it in the Oz web app

In the Oz web app, you can enable or disable Computer Use for:

New agent runs, when you start a run from the web app
Scheduled agent runs managed from the web app
Integrations, including Slack, Linear, and other integration-triggered agents

App settings, CLI flags, API config, and the web app are separate entry points. A scheduled or Slack-triggered agent does not pick up the desktop Experimental toggle.

Keep the sandbox from becoming a liability

Computer Use has unique security considerations. Warp notes that the risks are heightened when the agent interacts with the internet.

To minimize those risks:

Do not pass API keys, authentication tokens, or personal information to agents using Computer Use
If the environment has internet access, consider restricting it to an allowlist of known-safe domains
For tasks with real-world consequences, such as financial transactions or accepting legal terms, ask a human to confirm before the agent proceeds
Regularly review what agents are doing on your behalf, especially when testing new workflows

Isolation already blocks local credentials and local desktop access. Still treat prompts, tokens, and outbound domains as your responsibility.

Once Computer Use is on, Warp’s documented workflows include building from a Figma design or mockup and testing the result, visual regression after code changes, form and validation testing, responsive layout checks, and multi-step web flows. Agents can record sessions as annotated video—clicks, key presses, and cursor movement, with idle time trimmed out—and attach captured screenshots and recordings to pull request descriptions.

Choose the switch that matches your launcher

Start from the Warp app: flip Computer use in Cloud Agents under Experimental. Start from the Oz CLI: pass --computer-use or --no-computer-use. Start from the API: set config.computer_use_enabled. Start from the web app, a schedule, or an integration: configure Computer Use there. If the harness is Claude Code, Gemini, or Codex, assume Computer Use is off until you enable it.

Next step: Open Warp’s Computer Use docs and enable the path you actually use—app setting, Oz CLI flag, or config.computer_use_enabled—then run one UI-verification task and review the sandbox recording before you attach this to scheduled or integration-triggered agents.