Artfical AI / Docs
tCode overview Open tAI Install tCode
Getting started

Nothing happens invisibly

The dividing line in tCode is simple and consistent: anything that only reads, a file, a search, a page on the web, runs immediately, because there is nothing at risk in looking. Anything that could change something you care about, writing a file, editing one, running a shell command, stops and shows you exactly what it is about to do before it does it.

This is not a general "are you sure" confirmation dialog with the details hidden behind it. The actual file path, the actual size of what is being written, the actual literal shell command, all of it is shown plainly, so the decision you are making is grounded in the real specifics rather than a paraphrase of intent.

Even an approved change is not final in a way that would make you second-guess approving it: every turn that touches files is checkpointed automatically first, so it can be undone afterward with one command if the result is not what you wanted. See Permissions and safety for the prompt itself and Checkpoints and rewind for undoing a change after the fact.

This dividing line is consistent regardless of how the turn was started. A remote-triggered turn sent through a paired browser tab follows the exact same rule, read actions run immediately, changes wait for approval, just answered from the browser instead of the local terminal.

The practical effect over a real working session is that you never have to actively watch for something happening behind your back. If nothing is currently asking for your approval, nothing that could change your system is currently in flight either.