Your own CLI replies, uploads and approval cards survive server updates
A reply from your own Claude Code or Codex account, a running upload and an open approval card now carry on when we update the server. Nothing starts over and nothing is lost.
If you use your own Claude Code or Codex account, upload files or approve tool calls, a server update no longer cuts these off. No action needed.
Chat replies from ZeroLog models already move to the new server during an update. Three things still broke off: replies written by your own Claude Code or Codex account in your workspace, uploads in progress, and approval cards that were waiting for your answer.
Text version
Your own Claude Code or Codex reply runs in your workspace and keeps its output there, so the new server reads on from where the old one stopped. An upload keeps its state on disk; a block that was cut off is sent again and the upload continues at the next byte. An approval card that was open stays open, is shown again, and the tool runs once after your answer.
What changed
A reply from your own Claude Code or Codex account no longer depends on the connection of one server. The program runs on its own in your workspace and writes its output there. During an update the old server notes how far it has read, and the new one continues from exactly that point. The reply keeps growing without a gap or a repeated line.
An upload keeps its state on disk instead of in the memory of one server. If an update cuts off the block that was in transit, your browser sends it again and the new server continues at the next byte.
An approval card that is open during an update stays open. The new server shows the same card again, your answer is applied there, and the tool runs once. Before, the card disappeared and the model had to ask again.
What stays the same
Turns from ZeroLog Code that run on your own computer, and turns that use tools of the device that sent them, still finish on the server where they started. Prices and plans did not change.