Local folder or Docker sandbox for AI tasks?

Choose the environment according to the files and tools a task needs. Local-folder mode is convenient for an existing project; sandbox mode provides a deliberate file and process boundary with Docker.

Compare the practical differences

Decision	Local folder	Docker sandbox
Best fit	Existing host tools and project files	Dedicated task workspace and selected read-only resources
Useful property	Runs through the local CLI under your user permissions	Requires Docker and its prepared image
What to check	Local document dependencies must already be available	Can still contact model services over the network

A concrete choice

A small document task that needs the included document libraries may fit the prepared sandbox. A task depending on a specialized host tool may need a local project copy.

Verify the choice

Inspect the selected mode in Overview, run one small task and open its output. Do not describe a local folder as isolation or a network-enabled sandbox as offline.

Write the reason for your choice into the task brief or setup notes. Then run a small representative task and compare the actual result with the requirement that drove the decision. Keep any unmet requirement visible rather than declaring the whole workflow ready from a single successful step.

Continue with the setup

The related desktop guide explains the relevant controls and limits. Use the current application state when applying this comparison; the presence of an option in a document does not mean every dependency on your computer is ready.