OpenManus: Phone, schedules, and trade-offs

Other agents in this guide use their fourth page for phone chat and scheduled jobs, but OpenManus has neither, so this page lists what it does not do. It then weighs each setting it does have, using the same three questions as the rest of the guide.

Checked 2026-11-14.

What it does not do

• No phone. The README shows the terminal only, and its Feishu link is a chat group for developers, not a way to talk to your agent.
• No schedules. The README has no section on running jobs on a timer.
• No app or buttons. There is no desktop app, so you type everything.
• No yes-or-no before it acts. The README describes no approval step.
• No box. The README describes no sandbox, and the optional [sandbox] section is off in the example settings.
• No money limit. The 20-step limit is not a dollar cap.
• No update steps. The README does not explain how to update.

Trade-offs: safety versus capability

The [sandbox] section. The example settings file has a [sandbox] section that names a Docker setup, and it starts switched off with use_sandbox = false and its lines commented out.

• You gain: Nothing you can count on yet. The README does not explain the section, and this guide could not confirm what it covers or that it works.
• You risk: With it off, which is the default, any code OpenManus writes runs on your real computer as you. Turning it on without knowing what it covers could leave you trusting a wall that may not be there.
• Makes sense when: You can read the code and test it with Docker yourself. If you are a beginner, leave it as it is and rely on the rules in Keep it in a box.

The browser part. By default the browser part "attaches to Chrome or Chromium automatically," and setting OPENMANUS_DISABLE_BROWSER_USE=1 turns it off for the window you type it in.

• You gain: With the browser on, OpenManus can look things up and work on websites to reach your goal.
• You risk: Assume it uses your own Chrome, so any site you are logged into there could be within its reach, and a web page can hide instructions that the agent may obey.
• Makes sense when: You have signed out of email, banking, shopping, and calendar in Chrome, and you are watching the whole run. For your first jobs, keep it off.

Running with no approval step. OpenManus has no approval setting to turn on, because the README describes none, and its "ask the human" tool is something the AI may choose to use rather than a gate.

• You gain: A job runs from start to finish without interruptions, and each python main.py stops after one job or 20 steps.
• You risk: Nothing pauses before it deletes, moves, or overwrites a file. Your only brake is reading the lines as they scroll and pressing Control and C, and the 20-step limit stops loops, not spending.
• Makes sense when: The job stays inside agent-tryout, its worst case is something you can undo, and you are watching the whole time. It does not make sense for files you have not backed up, or for anything that touches your accounts or money.

For the three questions behind these choices, see Safety versus capability.

Sources

Opened 2026-11-14.

• OpenManus README (terminal only, Feishu community group, Browser Use, OPENMANUS_DISABLE_BROWSER_USE, no update steps): https://github.com/FoundationAgents/OpenManus/blob/main/README.md
• OpenManus example settings ([sandbox] block): https://github.com/FoundationAgents/OpenManus/blob/main/config/config.example.toml
• OpenManus main agent (tools, "ask the human", 20-step limit): https://github.com/FoundationAgents/OpenManus/blob/main/app/agent/manus.py
• This project's notes: other agents, safety checklist

Next

Continue to Playbook.