Trying to get started using the Antigravity CLI, located in Norway, running our coding environment on Hetzner hosted in region Finland.
When attempting the Gemini model in the Antigravity CLI i just get the messages below. When selecting Sonnet 4.6 as model everything works fine. Suggestions?
Agent execution terminated due to error.
Error ID: b942c9a7-5fde-47d9-808b-cf73b87ae333-2
────────────────────────────────────────────────────────────
test
Agent execution terminated due to error.
Error ID: b942c9a7-5fde-47d9-808b-cf73b87ae333-5
────────────────────────────────────────────────────────────
test
Agent execution terminated due to error.
Error ID: b942c9a7-5fde-47d9-808b-cf73b87ae333-8
Agent execution terminated due to error.
Error ID: b942c9a7-5fde-47d9-808b-cf73b87ae333-21
/model
⎿ Model set to Gemini 3.6 Flash (Medium)
────────────────────────────────────────────────────────────
can you see this input message?
Agent execution terminated due to error.
Error ID: b942c9a7-5fde-47d9-808b-cf73b87ae333-24
Best regards, AS
Same issue here, and I can confirm it is specific to the Gemini channel. Environment: macOS, latest Antigravity IDE + agy CLI 1.1.13, Google AI Pro subscription. Since ~Sep 9, every Gemini model (Gemini 3.8 / 3.7 / 3.1 flash variants) fails instantly with “Agent execution terminated due to error.” in both the IDE and the CLI, while non-Gemini models (GPT-OSS-120b on my side, Sonnet 4.6 per the OP) execute the exact same prompt fine. IDE language-server logs show the failing requests being routed to the canary endpoint daily-cloudcode-pa.googleapis.com, followed by repeated language_server_pb AcknowledgeCodeActionStep “run state not found” errors. No HTTP 429 involved, so this does not look like quota — more like a server-side issue on the Gemini routing/backend that Antigravity uses.
(Error ID example from my CLI runs: same format as OP, immediate termination after submit.)
More detail from server and logs:
- Server: Hetzner VPS location Finland, CentOS Stream 10, x86_64, KVM, 12 vCPUs, 22 GB RAM.
- System: kernel 6.12.0-218.el10.x86_64, glibc 2.39, Bash 5.2.26, tmux 3.6a, SELinux Enforcing.
- Client: native agy 1.1.27, installed September 8. Running over SSH inside tmux as a non-root user.
- Account: Google AI Pro with Google sign-in. I’m in Norway; Google’s Terms of Service page also shows Norway for my account when I go there to check.
- Network: no proxy or provider-override environment variables in the CLI process. September 8 checks classified both outbound IP families from VPS as Finland.
Found underlying error showing as:
FAILED_PRECONDITION (code 400): User location is not supported for the API use.
Logs confirm this for Gemini 3.8 Flash (High) on September 8. Gemini 3.1 Pro (Low) and Gemini 3.6 Flash (Medium) also failed on September 9.
Google sign-in and model listing succeed. Sonnet 4.6 test returned OK in 2.66 seconds. Gemini also failed from an empty workspace using a controlled IPv4 connection through
the same server address.
My logs also reference daily-cloudcode-pa.googleapis.com. However, the inspected conversation log contains neither run state not found nor AcknowledgeCodeActionStep errors. My observed rejection is HTTP 400. I cannot confirm that our reports share the same cause.
For support correlation: error b942c9a7-5fde-47d9-808b-cf73b87ae333-2 occurred on September 8, 2026, at 18:36:20 UTC.
Google’s availability FAQ ( FAQ | Google Antigravity Docs ) lists Norway and Finland. Could Google investigate which account/location eligibility check rejected these requests?
Version note: I still have 1.1.27. Google’s Linux release manifest now offers 1.2.0, which I haven’t tested yet. Will attempt that later.
Upgraded the Linux CLI from 1.1.27 to 1.2.0 and retested, Gemini 3.8 Flash (High), 3.1 Pro (Low), and 3.6 Flash (Medium) still return:
FAILED_PRECONDITION (code 400): User location is not supported for the API use.
Model listing still works, and Claude Sonnet 4.6 returns OK under the same conditions.