For anyone who has been experiencing the persistent “Agent terminated due to an error” issue over the last month — it is now fixed in Antigravity v1.22.2.
I want to share this update publicly so the community and dev team are aware, and to make the case that this regression should be documented and guarded against in future releases.
What happened
Starting around early March 2026, Antigravity began terminating agents mid-task with the error “Agent terminated due to an error.” The issue:
Affected all models, but was especially severe with Claude Opus — it would terminate repeatedly, sometimes within seconds of starting a task
Was not limited to complex tasks — even simple requests would trigger termination
Persisted across fresh conversations, workspace resets, and model switches
Made the tool essentially unusable for any sustained agentic work
Lasted approximately one full month
Why this was an IDE/platform bug, not a model issue
The critical evidence: a simple IDE update (v1.22.2) completely resolved the problem. If this were a model-side issue (context limits, token issues, model instability), updating the Antigravity client would not have fixed it. The fact that the fix came entirely from the client update confirms this was a bug in how Antigravity managed agent sessions, not in the underlying models.
Since updating to v1.22.2, I have had zero agent termination errors across extended, complex agentic sessions with multiple models including Opus.
Request to the dev team
Document the root cause — What specifically changed in the agent session management that caused this regression? What was the fix?
Add regression tests — A month-long stability regression in the core agent execution loop is serious. This should be caught by automated tests before it ships.
Communicate proactively — Many users likely abandoned or reduced usage during this period. A post-mortem or changelog note would rebuild trust.
For other users
If you were experiencing this issue — update to v1.22.2 and you should be good. If you’re still seeing terminations after updating, reply here so we can distinguish between the fixed bug and any remaining issues.
Thank you for this insightful feedback! We’re thrilled to hear that Antigravity is helping you build more efficiently and professionally. We appreciate you being part of the community and are excited to see your progress!
Update: Issue is partially back - needs further investigation
Following up on this post. jefferson_Lee is right - the issue has not fully resolved.
**Current state (as of April 9, 2026) - Before v1.22.2: Agent terminated every few seconds - completely unusable
After v1.22.2: Agent terminates 3-4 times per conversation before completing the task - still disruptive but functional New error message observed:
The termination now surfaces a different error: “Our servers are experiencing high traffic right now, please try again in a minute.” - but clicking Retry immediately brings the agent back in seconds. If this were genuine server load, retry after a few seconds would not work. This strongly suggests it is still a client-side bug masking itself as a traffic error, not actual server congestion.
Why this matters: 3-4 forced retries per conversation adds significant friction. For paid enterprise users on AI Ultra for Business, this is well below acceptable reliability. If this continues, users will migrate to alternatives (Codex, Claude Code, etc.) that don’t have this issue.
Updating my original PSA: the bug is only partially fixed. The v1.22.2 update reduced severity but did not fully resolve the root cause.
Hi all - i’ve had the “Agent terminated due to error” issue now for a while and FIXED IT.
Context: was really ONLY happening for me on Gemini 3.1 models. Claude worked fine. I downloaded the new version, same issue.
The problem was a bad tool call in my MCP server (mcp_config.json). Specifically, the Clickup MCP. This was passing into the API call and causing it to error out.
How i fixed it: i literally dumped the error logs into gemini (when you get the error, there’s a button to copy the error logs) and asked was going on, it told me there was a faulty tool call, most likely in the MCP config. A bit of back and forth, found the buggy one in Clickup and added it to the “disabledTools” array.
I hope this helps someone else. I know the fix may be different for you, but if you’re having this issue persiste, give this a shot.
- Before v1.22.2: Agent terminated every few seconds - completely unusable