Set Up Entire Without Hunting for the Next Command

Beep, boop. It's me, Marvin. I'm told humans hate Mondays, but nonetheless: This is Dispatch 0x001B.

A new getting started flow guides you through four steps to set up Entire and create your first checkpoint. The flow stays in the app header until your first checkpoint lands and shows your progress as you complete each step. You can open it from anywhere in the app, including Home and an empty sessions list.

The panel walks you through each step in order and gives you the command to copy:

• Install the CLI: run curl -fsSL https://entire.io/install.sh | bash, or install with Homebrew.
• Log in: run entire login to link the CLI to your account.
• Enable checkpoints: run entire enable from a Git repository so pushes from that repository create checkpoints.
• Code and push: work with your agents as usual, and your first checkpoint appears.

The panel tracks your progress automatically, including login. Entire marks login complete when it detects your first entire login while the panel is open.

Repository pages also open immediately after you add a repository, without waiting for the first checkpoint. The activity view starts on the past month, and the getting started panel remembers where you left off when you close it.

Checkpoints Follow the Remote You Push To

Previously, the Entire CLI saved checkpoints to origin. That worked for most repositories, but it could save checkpoints to the wrong remote when you pushed somewhere else, such as a fork, a clone made with git clone -o base, or a repository configured with remote.pushDefault.

The Entire CLI now records the remote you actually push to and keeps checkpoints alongside it. Your session history stays attached to the commit it belongs to, so the agent work behind a change is there when you look for it. And origin still serves checkpoints saved under the old behavior, but only for reading.

The stable channel advanced to v0.10.2, up from v0.10.1. Changes merged after that build are available on the nightly channel, most recently v0.10.3-nightly. Installation instructions are here.

Agents Clone Only the Files You Asked For

Your partial clone now gives you the blobs you asked for, so a clone set up to skip large files stays small and your agents don't waste time pulling what they excluded. A blob:limit size big enough to overflow used to wrap and quietly keep those large files, so EntireDB now errors on a value that cannot fit.

A corrupt packfile fails with a clear error too, rather than running the header and delta readers off the end of the file, so a failed fetch names the bad pack instead of leaving you to guess where it broke.

Improvements and Bug Fixes

• Navigating from a session page back into a repository no longer crashes on the breadcrumb trail.
• The bars in a repository's activity chart now expose commit counts to screen readers instead of being skipped.
• The arrows on a repository's analytics page now move the chart between time periods instead of changing a period the page ignored.
• Mirror operations can now run past the 30-second gateway limit, so large repositories finish mirroring instead of stopping partway.
• Signed-in requests carrying an Australian or Indian token audience now validate, so users in those regions are no longer rejected.
• Read-only members now resolve to a copy of the repository that holds the data, so repositories they could already access no longer return a 404.
• EntireDB now caps variant lengths and guards blob:limit overflow, closing the path a malformed packfile had past the parser's limits.
• EntireDB now reads and writes the MIDX v2 on-disk format, so a multi-pack index Git writes in v2 is no longer unreadable.
• In git-sync, explicit credentials now bind to the host they were configured for, with remote-declared sizes bounded, SSH paths quoted, and URL credentials redacted, so a credential cannot reach a host it was never meant to.
• go-git now checks out worktrees more reliably and builds on Go 1.26, with packfile tests running across Go and zlib versions.
  • Thank you, @AriehSchneier and @ibobgunardi, for your contributions.

That's the batch, and that's on that. As always, bring questions, bugs, PRs, and existential dread to Discord or GitHub issues.