Since ~midday today, git sync for this app is failing two ways: (1) the in-editor AI assistant's git pull main / "Integrate changes" fails with "Tool execution was interrupted before completing"; (2) the MCP/AI "new build from main" provisions an empty starter sandbox (no app code, git log shows only "Newly created app" + "[auto] Sync sandbox setup") instead of cloning current main. Existing editor/threads have the code but are pinned to a stale origin/main and cannot fetch newer commits. The published app itself works fine. Looks like the app's git remote/main can't be fetched or pulled into any working copy. Please investigate the source-control backing for this app.
Welcome to the community, @poisonik! The majority of the backend errors that I see are actually related to Vite port collisions, not source control. The fix for that particular issue went live in version 4.17.0 - can you let me know if you're still having issues?
You might need to try initializing a new thread via the UI:
Thanks @darren! A few clarifications from our side:
We're on Retool Cloud (org squadtalentxyz), not self-hosted — so there's no 4.17.0 version for us to set or check. Can you confirm whether the Vite-port-collision fix has been rolled out to our Cloud instance, and if not, when it will be?
We're still hitting this today (tested repeatedly over several days). It's specifically through the MCP / "build app with AI" integration on our app "Candidate Triage Dashboard" (UUID 7a563906-6ace-11f1-b9da-13213322658e). Symptoms:
Starting a new build thread from main provisions an empty scaffold app — the sandbox has none of our code (/backend/triage and /frontend/pages don't exist), and git log shows only Newly created app + [auto] Sync sandbox setup. The repo is never cloned into the sandbox. Each attempt also creates a throwaway "Untitled-N" app.
Runs frequently detach with zero commits shortly after starting.
Reviving an older existing thread doesn't help — git pull reports "Already up to date" but HEAD is stuck at a stale commit; it can't fetch newer commits from main, so it can't reach our current published version.
What does work: continuing a freshly initialized in-editor thread (created from the editor UI) — it has our real code and can build/publish. But its live sandbox connection drops as soon as the editor tab loses focus, which kills in-progress runs.
If all of this traces back to the Vite port-collision crash (sandbox dying at startup before it clones the repo), confirming the fix is live for our Cloud org should unblock it. If it's a separate seeding/source-control issue, please advise. Happy to provide thread/generation IDs or timestamps of failed runs.
Hmm thanks for the clarification. Version 4.17.0 would have finished its Cloud rollout by last Thursday (7/2), so I certainly expect the Vite fix to have been live as of Monday. I don't see any of the same errors during your most recent MCP session, though, which is promising. It actually looks like you successfully published - can you confirm?
I think most of the additional debugging that I'd like to do would be better done on a live call. Are you able to join our scheduled Office Hours?
The fix for that particular issue went live in version 