Product: Google Antigravity CLI
CLI version: 1.1.10
OS: macOS
Network egress observed: Singapore, AS21859 (Zenlayer Inc)
Account authentication: OAuth succeeds normally
Issue:
Gemini requests in Antigravity CLI intermittently fail with:
FAILED_PRECONDITION (code 400): User location is not supported for the API use.
The CLI UI only shows:
Agent execution terminated due to error.
This is not a stable account or network failure. The same authenticated CLI process, same machine, and same network
can alternate between successful streaming responses and the 400 location error within seconds.
Reproduction:
- Start agy and authenticate successfully.
- Select Gemini 3.6 Flash (High).
- Send a trivial prompt such as: hi
- The request frequently fails before generation with the location 400 error.
- In the same session, switch to Claude Sonnet 4.6 (Thinking).
- Claude succeeds repeatedly.
- Switch back to Gemini 3.6 Flash (High); the location 400 returns almost immediately.
Evidence from one session:
- Between 16:53:44 and 17:15:20, with Claude Sonnet 4.6 selected:
47 successful streamGenerateContent responses, 0 location errors. - After switching back to Gemini 3.6 Flash at 17:15:25:
5 requests failed with the location error within about 35 seconds, while only 2 streaming requests succeeded. - Both IPv4 and IPv6 egress checks report the same location/provider: Singapore / AS21859 Zenlayer Inc.
- OAuth authentication succeeds before these failures.
This strongly suggests inconsistent region/IP eligibility handling in the Gemini/Antigravity backend path, rather
than an invalid local authentication state. The same account also works from another Windows machine.
Example CLI Error IDs:
- 5e406aa4-696f-49c0-bea0-7a84336fff1b-3
- 5e406aa4-696f-49c0-bea0-7a84336fff1b-7
Please investigate why the Gemini route intermittently classifies the same authenticated client and egress as
unsupported, while the Claude route remains stable.