Having the same problem since weeks ..
Always the “Agent terminated …” and I tried all the tricks listed in the field. Seriously I wonder what I should stick with my Gemini subscription if I can’t use this tool that integrates it so deep that it had be far the best user experience and helped my the past months - now I am stuck with a piece of broken software.
Fix that please !
is this still unresolved? i might have to just go to cursor or windsurf at this point.
still not working and i don’t know what to do
With all due respect, your statement is nothing more than a way to buy time, because you clearly don’t know what’s happening.
Let me be clear, Abhijit, this isn’t a message addressed to you as a representative of the organization. It’s a direct criticism of the company. It’s been eight days since your last message, and the situation remains the same.
I can’t believe that a big tech company like Google doesn’t have the ability to fix a bug that has been going on for weeks in a product that has literally just been launched on the market. It’s only logical that we have high expectations for this IDE, but apparently our money isn’t worth the effort on the part of the developers.
And do you know what the result of this behavior on Google’s part is? We customers are looking for other solutions. I’ve already canceled my AI plan and purchased a plan for Claude Code.
Sorry, but I can’t waste time with incomplete software that’s nothing more than a beta version.
Anyway, best of luck to all the people who still believe and pay for this project.
Two full weeks. That is how long I have been paying for AI Ultra without being able to use the service for a single minute. 14 days of HTTP 500 errors and 0% uptime. Support keeps telling me ‘engineers are investigating’, but no actual ETA or workaround is provided. Is anyone else still stuck in this nightmare?
I’m in the same boat. Welcome to the club. Google support keeps claiming they need more time, but I’ve seen zero improvement. Considering I’m on AI Ultra and paying a premium, this is pretty disappointing.
Huh, I solved my problem by disabling one of my mcp.
Users of the Antigravity editor (or any Gemini-powered agent) experience a complete failure of the AI assistant when the @a-bonus/google-docs-mcp server is active. The assistant returns a 400 Bad Request error instead of responding to prompts.
The error is caused by a JSON Schema validation failure within the Google Gemini API (GenerateContentRequest).
Gemini’s API enforces strict schema validation for tools (function calling). Specifically, any parameter defined as an array must have an items property. In the case of 2D arrays (common in Google Sheets for rows and columns), the schema must define items for the outer array and another items property for the inner array.
The @a-bonus/google-docs-mcp package uses z.any() for the innermost values of its spreadsheet tools. This causes the generated JSON schema to look like this:
"values": {
"type": "array",
"items": {
"type": "array"
// MISSING "items" property here!
}
}
Gemini rejects this with the error: items.items: missing field.
The following tools in the @a-bonus/google-docs-mcp package (located in the local npm cache) are responsible for the crash:
| Tool Name | Affected Parameter | File Path (Relative to package root) |
| :— | :— | :— |
| writeSpreadsheet | values | dist/tools/sheets/writeSpreadsheet.js |
| batchWrite | data[].values | dist/tools/sheets/batchWrite.js |
| appendSpreadsheetRows| values | dist/tools/sheets/appendSpreadsheetRows.js |
| createSpreadsheet | initialData | dist/tools/sheets/createSpreadsheet.js |
The z.any() definition must be replaced with a more explicit type that Gemini can validate, such as a union of primitives (string, number, boolean, or null).
Current Code:
values: z.array(z.array(z.any()))
Corrected Code (Gemini-Compatible):
values: z.array(z.array(z.union([z.string(), z.number(), z.boolean(), z.null()])))
Same issue here. Can´t use my Antigravity IDE with a Pro Account since weeks now, which renders it useless. I sent tons of bug and Support requests to google, but never got an answer. Thinking of switching to claude code now..
seriously..; “11 chat messages at model retry attempt 1 and API retry attempt 1
2026-03-07 15:52:31.972 [info] E0307 15:52:31.971286 14196 log.go:380] agent executor error: INTERNAL (code 500): Internal error encountered.: INTERNAL (code 500): Internal error encountered.
2026-03-07 15:52:33.795 [info] I0307 15:52:33.795059 14196 planner_generator.go:288] Requesting planner with 13 chat messages at model retry attempt 1 and API retry attempt 1
2026-03-07 15:52:34.644 [info] E0307 15:52:34.644127 14196 log.go:380] agent executor error: INTERNAL (code 500): Internal error encountered.: INTERNAL (code 500): Internal error encountered.
2026-03-07 15:52:35.906 [info] I0307 15:52:35.905727 14196 planner_generator.go:288] Requesting planner with 15 chat messages at model retry attempt 1 and API retry attempt 1
2026-03-07 15:52:36.831 [info] E0307 15:52:36.830830 14196 log.go:380] agent executor error: INTERNAL (code 500): Internal error encountered.: INTERNAL (code 500): Internal error encountered.“
im going to be on this board untill Im old and grey, an various users are right, I get google is a massive company but surely someone could allocated even just an LLM agent to handle internal external communication to update people as to what is going, out it on a background process priority i guarantee it will get more done than an actual google employee at this point…no response to antigraviy submitted error logs, no response to users on forum posts… come on google pull your socks up man..
It’s been almost three weeks since I abandoned Antigravity in favor of Claude Code. Today we decided to check if they’d fixed the 500 error… I don’t know what I was expecting, the account is still unusable, all models return an error, and the logs modestly mention only Server Error.
Same for me : it doesn’t work since more than 3 weeks. By the meantime I cancelled my subscribtion. Since a few day I’m back on a free plan… and it doesn’t change anything. Agent mode is still broken on that account.
Yep, just cancelled subscription here too.
For the life of me I do not understand what google is doing at the moment. First AIStudio gets destroyed - fine, I thought, I’ll pay up and move to Antigravity.
And now AG is useless because none of the models respond (while still eating away at quota ofc.).
I’m done with this.
Good luck Google, you’re going to need it.
A workaround for me is to have installed the [Antigravity ToolKit extension]datafrog..io/es /docs/toolkit-for-antigravity). Then, just before asking Gemini to do something, I hit Restart in the extension, wait like 1-2 seconds after the restart is complete, and then Retry the discussion. This is annoying, but works every time.
Thank you for bringing these concerns to our attention. Please be assured that I have shared your feedback with our internal team for further review.
We appreciate your continued patience as we work to enhance the Antigravity experience.
(post deleted by author)
Have same issue with Pro, The error
Trajectory ID: 664f0f6b-3f48-409d-941b-d4abd74…
Error: agent executor error: could not convert a single message before hitting truncation
(1) attached stack trace
– stack trace:
| google3/third_party/jetski/cortex/cortex.(*CascadeManager).executeHelper.func1
| third_party/jetski/cortex/cascade_manager.go:2096
| […repeated from below…]
Wraps: (2) agent executor error
Wraps: (3) attached stack trace
– stack trace:
| google3/third_party/jetski/cortex/chatconverters/chatconverters.init
| third_party/jetski/cortex/chatconverters/trajectory_chat_converter.go:44
| runtime.doInit1
| third_party/go/gc/src/runtime/proc.go:8105
| runtime.doInit
| third_party/go/gc/src/runtime/proc.go:8072
| runtime.main
| third_party/go/gc/src/runtime/proc.go:258
| runtime.goexit
| third_party/go/gc/src/runtime/asm_arm64.s:1447
Wraps: (4) could not convert a single message before hitting truncation
Error types: (1) *withstack.withStack (2) *errutil.withPrefix (3) *withstack.withStack (4) *errutil.leafError
Just a quick update for everyone following this saga: My case has now been ‘handed over to a specialist’ for the second time in three days.
20 days of 0% uptime on a paid AI Ultra subscription,Still getting the same HTTP 500 error. Still no workaround. Still no refund.
I saw the update and I gave Antigravity a spin. Now it seems even worse!
I tried a few minutes ago with my personal account: six “error code 500” in a row trying to use Claude Opus.
Result: no response and quota ran out without even analyzing the .cpp file I gave to it.
The only LLM that works is Gemini Flash (useless for verifying code).
Speechless.
yeah i noticed that as well, the pro high and low seemed to be the only versions able to verify decent code, flash just mangles thing..
and my main account is still giving internal errors.. this product is a joke
On paper, Gemini 3.1 Pro High should beat Claude Code Opus 4.5 and even 4.6 (which tends to produce more elegant code but is slower) when it comes to programming. In reality, the difference in precision and optimization is staggering. Even Claude Sonnet 4.6 is significantly more efficient than Gemini for these types of tasks.
Gemini Flash versions are excellent for documentation where tinkering isn’t much needed: grammar, construction, and optimization.
For brainstorming, however, Gemini is definitely better than any version of Claude.
Different LLMs for different tasks, I suppose…