Checkpoints & Undo

When the assistant edits files, you get a way back. Boltise takes a checkpoint at the start of every turn — every message you send — and you can restore to any of them.

Restoring

Hover any of your own messages in the chat and choose Restore checkpoint. Two options:

Option	What happens
Restore files	Your files go back to their state just before that message ran. The conversation stays.
Restore & edit	The same, and your message is put back in the composer so you can rephrase it and try again.

Restore & edit is the one you want after a request that was misunderstood. Rather than arguing with the result, you rewind and ask better.

What it does and does not cover

Restoring reverts files on disk. It does not:

restart or roll back your running server — restart it yourself if the reverted code is already loaded
undo database writes — a checkpoint cannot un-run an INSERT
undo anything that left your machine

How it differs from Ctrl+Z

Ctrl+Z is per-file and per-editor: it undoes your typing in the file you are looking at. A checkpoint is per-conversation-turn and covers every file the assistant touched in that turn, which is usually several and often files you never opened.

They are also stored differently. Editor undo lives in memory and dies with the tab; checkpoints are written to disk as the edits happen, so they survive closing the file, closing the project, and restarting Boltise.

Reverting one file instead

If a single edit was wrong and the rest of the turn was fine, you do not need a checkpoint — ask the assistant to revert that one file. It goes back to its state at the start of the conversation, leaving everything else alone.

After restoring

Restoring is itself final: the checkpoints you rewound past are discarded, and they stay gone after a reload. You cannot restore forward again, so if you are unsure which checkpoint you want, start with the most recent one and work back.

Docs privacy