Goose

Goose: Phone and schedules

This is page 4 of 8 for Goose. It covers the two ways an agent can work while you are not typing to it, which are schedules and phone chat. Then it weighs the main Goose settings, so you can decide how far to loosen each one.

Later: schedules (and why not your phone)

Schedules

Goose can run a saved job on a timer. In the desktop app, open the Scheduler view, or start from the Recipe Library. A recipe is a saved set of instructions, and the playbook walks you through making one.

When you set a schedule, you pick a name, the recipe, how it runs, and how often. "Background" runs with no window and saves the results, while "Foreground" opens a window if Goose is running.

This guide could not confirm whether a schedule still runs while the computer is asleep or you are logged out, so plan on leaving the computer on and logged in. Scheduled runs skip approvals even when Manual Approval is on (see the Scheduler trade-off below), so only schedule a job after it has worked safely by hand.

Why not your phone

There is no Goose phone app in the install guide, and the phone paths that once existed are mostly gone:

The docs say phone access through a secure tunnel "is no longer available in current goose Desktop builds," so that route is closed.
The goose Mobile page is marked archived.
The only phone path left is an experimental Telegram gateway. You set it up by typing commands, it only answers while your computer is awake and online, and the docs do not name a way to block strangers.

If texting your agent from your phone matters to you, Goose is the wrong pick. The main guide's phone and autopilot step uses Agent Zero.

Trade-offs: safety versus capability

Goose runs on your real computer as you, with no container around it, so each of these choices decides how far a single mistake can reach.

Default Mode. This setting, under Settings, then Chat, then Default Mode, decides whether Goose asks before it acts. The choices are Completely Autonomous, Manual Approval, Smart Approval, and Chat Only, and a chat keeps the mode it started with, so start a new chat after you change it.

You gain: Completely Autonomous lets Goose run commands and edit, move, or delete files without stopping, so long jobs finish with no clicks at all. Smart Approval approves the actions it rates as low risk, which cuts down the clicks without removing them. Chat Only goes the other way, with no extensions and no file changes, which is the safest choice but leaves you with a chatbot.
You risk: in Completely Autonomous, the docs say Goose "can modify files, use extensions, and delete files without requiring approval," and on your computer those are your real files. Even Manual Approval asks only for actions the model classifies as writes, and the docs call that sorting a "best effort" by the model.
Makes sense when: you are following this guide, because Manual Approval is the right default for every job in it, and Chat Only suits a day when you only want to talk. Completely Autonomous does not make sense for a beginner, because without a container or a folder lock, an unwatched mistake can reach anything your account can.

Where Goose can reach. Goose has no folder lock, so the folder you name in your message is a request, not a wall. Every command it runs can reach any file your everyday account can reach.

You gain: Goose can work on real files wherever they already live, without you copying them into a special folder first.
You risk: one wrong or tricked command can delete or overwrite files far outside the folder you named, and the only thing that stops it is you reading each Allow request.
Makes sense when: you are starting out and work in the empty goose-tryout folder, or later, when a real folder is backed up and Manual Approval is on. If you want a real wall, the safety research lists an optional terminal command, goose session --container <container_id>, that runs extensions inside a Docker container, and Advanced features explains what it needs and why it is not a beginner step.

Extensions. Extensions give Goose its tools. The built-in Developer extension, which runs commands and manages files, is on by default, and full browser control is a separate skill you install.

You gain: each extension or skill adds a new kind of job, such as clicking around websites for you instead of only searching them.
You risk: each one is one more way for a mistake or a prompt injection to act for you, and if a browser it drives is signed in to your accounts, it can act on those sites. The docs say Goose checks outside extensions for known malware and blocks a package it finds is malicious, but prompt-injection detection starts off.
Makes sense when: a job truly needs the new tool, you add one extension at a time, and you stay logged out of email, your bank, and stores in any browser Goose can drive. If a job needs no tools at all, Chat Only turns all extension use off.

The Scheduler. The Scheduler view runs a saved recipe on a timer, and a "Background" run opens no window while it works.

You gain: a job, such as a regular summary, gets done without you starting it each time.
You risk: a scheduled run does not ask at all. A source check in a GitHub issue thread found that Goose's scheduler forces Completely Autonomous mode from version 1.46.0 on, whatever your Default Mode is, and ignores a mode written into the recipe. The official docs do not mention this. A repeating job also repeats any mistake until you notice.
Makes sense when: the recipe has already worked safely by hand, and it is meant to touch only its own folder (Goose cannot enforce that). You check the results after the first few runs, and you treat every scheduled run as running with no approvals. Do not schedule anything that deletes, moves, or sends.

To weigh these choices for your own jobs, use the three questions in Safety versus capability.

Sources

Checked 14 November 2026.

Next: Eight jobs for Goose.