The “agent terminated” error on Antigravity follows a clear daily timing pattern that I have not seen anyone break down specifically.
The window: From roughly 1 AM PKT (8 PM UTC) through early morning, the agent runs normally. Outside that window, from morning through the rest of the day until around 1 AM, the agent terminates almost instantly. Not after a few tool calls. Not mid-task. Within seconds of starting.
The progression over the past week:
About 4-5 days ago (around April 13-14), the situation was a complete blackout. The agent would not run at any time of day. 24/7 instant termination. No usable window existed at all. I could not complete a single task regardless of what time I tried.
Then something shifted. The late-night window opened up. From roughly 1 AM PKT onward, the agent started working. During that window, it runs fine. Normal behavior, completes tasks, no random terminations. But the moment you try to use it during standard working hours (say 9 AM to midnight PKT), it is back to instant termination. The pattern has been consistent every day since.
So the situation has improved from “completely unusable” to “usable for a few hours in the middle of the night.” That is progress, but it is not a functioning product. If the only reliable window for a paid tool is 1 AM to 6 AM in your timezone, that is a capacity problem being masked as an error.
What this pattern suggests:
This is not a client-side issue. It is not MCP conflicts. It is not context size. The same account, same machine, same workspace, same prompts work perfectly at 2 AM and fail instantly at 2 PM. The only variable is server load. The infrastructure cannot handle daytime traffic, and the “agent terminated” error is what users see when capacity runs out.
On a separate note: Claude Opus 4.7 shipped on April 16. Cursor and Claude Code had it on day one. It has been over 48 hours and Antigravity still does not list it. When Opus 4.6 launched, Antigravity picked it up almost immediately. The delay on 4.7 is a departure from that pattern and there has been no communication about it.
The ask:
- Acknowledge that the “agent terminated” error correlates with server load and time-of-day demand, not user-side factors.
-
- Publish a capacity roadmap or at minimum a status page so users can plan their work around actual availability instead of guessing.
-
- Provide an update on Claude Opus 4.7 availability in Antigravity.