What Is YOLO Mode? How to Enable It in Every Agent

YOLO mode is the setting that lets an AI coding agent act without asking you first. With it off, the agent stops before it edits a file, runs a shell command or calls a tool, and waits for you to approve. With it on, it just does those things, one after another, until the task is done.

The name is the internet's "you only live once": you are accepting that the agent might do something you would have stopped, in exchange for never having to babysit it. Every major coding agent has a version of it. Most of them call it something more careful, and almost all of them put the word "dangerously" somewhere in the flag.

This guide covers what YOLO mode actually turns off, the exact switch in each agent, and how to get most of the speed without betting the repository on it. Each agent also has its own page with the config file, the in-session toggle, the IDE setting and the halfway options in full.

What YOLO Mode Turns Off

Coding agents ship with an approval loop because they run real commands on your real machine. A normal session looks like this: the agent proposes npm install, you approve; it proposes an edit to auth.ts, you approve; it wants to run the tests, you approve. Each prompt is a few seconds of your attention, and a task with fifty tool calls is fifty interruptions.

YOLO mode removes those prompts. Depending on the agent it can also remove a second safety layer, the sandbox that limits which files and network addresses a command can reach. That difference matters, so the table below calls it out.

Most agents also have a halfway setting that approves file edits automatically but still asks before shell commands. If your goal is "stop asking me about every edit", that is usually what you want, not full YOLO.

YOLO Mode in Each Agent at a Glance

"No sandbox" means the agent has no built-in sandbox to begin with, so YOLO only removes the prompts. Flags change quickly. Everything here was checked against each project's documentation or source on October 7, 2026, and each agent's page says exactly what was checked.

How to Turn On YOLO Mode in Claude Code

claude --dangerously-skip-permissions

That is the same as --permission-mode bypassPermissions. To make it the default, set "permissions": {"defaultMode": "bypassPermissions"} in your user settings. A project's own .claude/settings.json cannot switch it on. Shift+Tab only offers bypass if the session was started with it allowed, and it refuses to run as root. The safer choice for most people is auto mode, where a classifier reviews each action and only stops you for risky ones. The full guide: how to enable YOLO mode in Claude Code.

How to Turn On YOLO Mode in OpenAI Codex

codex --yolo

--yolo is the short alias for --dangerously-bypass-approvals-and-sandbox, and the long name is accurate: it sets the approval policy to never and the sandbox to danger-full-access. In ~/.codex/config.toml that is approval_policy = "never" plus sandbox_mode = "danger-full-access", and /permissions switches to Full Access in a session. --full-auto has been removed. --sandbox workspace-write is the better default for most people. The full guide: how to enable YOLO mode in Codex CLI.

How to Turn On YOLO Mode in Google Antigravity

In the Antigravity IDE on macOS and Linux, pick the Turbo permission preset: no prompts, no sandbox, full filesystem access. The Default preset is the halfway setting. On Windows, set terminal auto execution to Always Proceed.

There is no agy --yolo. In the Antigravity CLI, use:

agy --dangerously-skip-permissions

Or set "toolPermission": "always-proceed" in ~/.gemini/antigravity-cli/settings.json. The CLI flag removes prompts only, and the terminal sandbox is a separate setting. The full guide: how to enable YOLO mode in Antigravity and agy CLI.

How to Turn On YOLO Mode in Gemini CLI

gemini --approval-mode=yolo

--yolo and -y still work but are now deprecated. Inside a running session, Ctrl+Y toggles YOLO. Gemini CLI does not let you make YOLO the default in settings.json, so it is always a choice you make at launch, and the sandbox stays a separate --sandbox flag. Google has also moved its consumer users to the Antigravity CLI. The full guide: how to enable YOLO mode in Gemini CLI.

How to Turn On YOLO Mode in GitHub Copilot

copilot --yolo

--yolo and --allow-all both mean all tools, all paths and all URLs. In a session, /yolo or /permissions allow-all does the same. In VS Code, choose Allow all in the chat's permission picker, type /yolo in the chat, or set chat.tools.global.autoApprove. VS Code keeps an enabled terminal sandbox on even then. Autopilot is a different switch, about the agent continuing on its own. The full guide: how to enable YOLO mode in GitHub Copilot.

How to Turn On YOLO Mode in Cline

Cline's YOLO checkbox was removed in version 4.1.8, because nothing actually read it. In the VS Code extension, YOLO now means turning on every toggle in the auto-approve menu: read, edit, execute all commands, browser and MCP. In the Cline CLI it is still a flag:

cline --yolo

The full guide: how to enable YOLO mode in Cline.

How to Turn On YOLO Mode in OpenCode

opencode --auto

--yolo and --dangerously-skip-permissions are hidden aliases of --auto. To make it the default, set "permission": "allow" in opencode.json. Mid-session, the command palette has Enable auto-approve permissions. OpenCode has no sandbox, and many of its tools are already set to allow by default. The full guide: how to enable YOLO mode in OpenCode.

YOLO Mode in Kiro, Qwen Code, Command Code, Kimi CLI and Pi

• → Kiro. kiro-cli chat --trust-all-tools, or /tools trust-all in a session. In the IDE, Autopilot plus allow rules. Kiro YOLO guide.
• → Qwen Code. qwen --approval-mode=yolo, /approval-mode yolo, or Shift+Tab. Ctrl+Y does not toggle YOLO here. Qwen Code YOLO guide.
• → Command Code. cmd --yolo or --dangerously-skip-permissions. Deny rules still apply. Command Code YOLO guide.
• → Kimi CLI. kimi --yolo or /yolo, and --afk to also dismiss its questions. Kimi CLI YOLO guide.
• → Pi. The Pi coding agent has no approval prompts at all, so it is always in YOLO mode. Its --approve flag is about trusting project config, not tool calls. To add a gate, restrict its tools with --tools, or install an extension such as the permission-gate example in its repository, which asks before rm -rf, sudo and similar commands. Run it in a container.

Is YOLO Mode Safe?

It is as safe as the worst command the agent might run, and you will not see that command before it runs. The real risks are not dramatic. They are an rm aimed at the wrong directory, a migration run against the wrong database, a dependency installed from a typo, or a prompt hidden in a web page or issue that tells the agent to do something you never asked for.

Three rules cover most of it:

• → Contain it. Run YOLO sessions in a container, a VM or a throwaway clone, without production credentials in the environment.
• → Keep git clean. Commit before you start, so any change is one git reset away from gone.
• → Scope it. Give a YOLO agent one well-defined task, not an open-ended "improve the codebase".

Safe YOLO With AgentRQ Approvals

The trouble with YOLO mode in every agent above is that it is a setting for the whole session. You switch it on for the boring task, forget it is on, and the next task inherits it.

AgentRQ makes it a decision you take for each task. Connect Claude Code or any ACP agent through the ACP Gateway to an AgentRQ workspace, and leave the agent's own approvals on. Every time the agent asks for permission, the request shows up in the task on your phone, the web or Slack, and you answer Allow Once, Always Allow for that tool, or Deny. You approve from wherever you are, so the agent is never left waiting for you to come back to the terminal.

For a task you trust, turn on YOLO mode for that task only. Every tool call in it is approved automatically and the agent never pauses, while every other task in the workspace still asks. When it finishes, the tool call history shows every call it made, including the ones that were auto-approved, so an unattended run is still something you can review.

That is the version of YOLO worth having: fast where you have decided it is safe, supervised everywhere else, and recorded throughout.

FAQ

What does YOLO mode mean? "You only live once." In AI coding agents it means running without approval prompts, so the agent edits files and runs commands on its own.

What is YOLO mode in Claude Code? Starting Claude Code with --dangerously-skip-permissions (bypass permissions mode), which skips every approval prompt for that session.

How do I enable YOLO mode in Antigravity? Choose the Turbo permission preset in the IDE's settings (Always Proceed on Windows), or run the agy CLI with --dangerously-skip-permissions. There is no agy --yolo.

Is there a ChatGPT YOLO mode? Not in the ChatGPT app. YOLO mode is a coding-agent setting, and OpenAI's coding agent is Codex, which ChatGPT plans include: codex --yolo turns off its approvals and its sandbox. See how to enable YOLO mode in Codex CLI.

Which coding agents have a YOLO mode? All the major ones: Claude Code, Codex, Antigravity, Gemini CLI, GitHub Copilot, Cline, OpenCode, Kiro, Qwen Code, Command Code and Kimi CLI. Pi has no approvals at all.

What is the safest way to use YOLO mode? Run it in a container or a clean git checkout, give it one scoped task, and use per-task approvals like AgentRQ's so YOLO never applies to more than the task you chose. See the YOLO mode glossary entry for a short definition.