Skip to content

Antigravity 1.5 quota exhaustion shows "Reset time unavailable; retry manually" despite JSON reset timestamp #40434

Closed as not planned

Description

@marcinsulikowski

What happened

During a normal Antigravity turn (gemini-4-neon), the account hit its 5-hour individual quota. The thread rendered two conflicting notices right next to each other:

  1. Assistant message from Antigravity:

    ⚠ Individual quota reached. Please upgrade your subscription to increase your limits. Resets in 2h10m40s.

  2. Yellow recovery banner directly below it:

    Usage limit reached — Reset time unavailable; retry manually

At the same time, Usage → Limits accurately showed Gemini · 5-hour at 0% left with a live reset countdown (1h 53m). Because the banner had no reset time, T3 Code could not display the reset time or schedule auto-resume.

Diagnosis

apps/server/src/provider/acp/AntigravityTurnOutcome.ts recognizes Antigravity 1.5's 429 RESOURCE_EXHAUSTED error header, but its reset-time extractors still only match Antigravity 1.4's unquoted Go map[...] format:

  1. watchAntigravityPromptOutcome reads the turn's final step_type = 17 (ERROR_STEP_TYPE) row from <stateDir>/providers/antigravity/<instance>/antigravity-acp/conversations/<sessionId>.db.

  2. In Antigravity 1.5.14, that step begins with RESOURCE_EXHAUSTED (code 429): Individual quota reached... followed by a JSON error body containing "reason": "QUOTA_EXHAUSTED", "quotaResetTimeStamp": "2026-11-21T16:26:59Z", "quotaResetDelay": "2h10m40.817699543s", and "retryDelay": "7840.817699543s".

  3. QUOTA_EXHAUSTED (/\breason:QUOTA_EXHAUSTED\b/) does not match JSON "reason": "QUOTA_EXHAUSTED", so turnFailure falls through to STATUS_ERROR (RESOURCE_EXHAUSTED (code 429): ...), which classifies the failure as class: "usage_limit" and calls resetAtFrom(text, now).

  4. resetAtFrom checks only three unquoted Go-map patterns:

    • RESET_TIMESTAMP = /\bquotaResetTimeStamp:([^\s\]]+)/
    • RESET_DELAY = /\bquotaResetDelay:((?:\d+(?:\.\d+)?(?:ns|us|µs|μs|ms|h|m|s))+)(?=[\s\]])/
    • RETRY_DELAY = /\bretryDelay:((?:\d+(?:\.\d+)?(?:ns|us|µs|μs|ms|h|m|s))+)(?=[\s\]])/

    None of these match JSON "key": "value" syntax, so resetAtFrom returns undefined and failure.resetAt is stored as null / omitted.

  5. packages/shared/src/orchestrationV2ThreadError.ts (threadErrorSummary) forwards usageLimitResetAt: null to apps/web/src/components/chat/UsageLimitRecoveryBanner.tsx (usageLimitRecoveryBannerItem), which falls back to "Reset time unavailable; retry manually".

  6. Usage → Limits still shows the countdown because apps/server/src/provider/antigravityUsageLimits.ts queries v1internal:retrieveUserQuotaSummary / v1internal:fetchAvailableModels independently and reads resetTime directly from the API.

Verified against main (efea1aab): AntigravityTurnOutcome.ts is unchanged since v0.1.12-nightly.20261121.5038 (9f9ebf61).

Steps to reproduce

  1. Run T3 Code desktop AppImage (0.1.12-nightly.20261121.5038) with managed Antigravity ACP 1.5.14 signed in via oauth-personal.
  2. Run a thread on gemini-4-neon (high) until the account exhausts its 5-hour individual quota mid-prompt.
  3. Observe:
    • Chat displays ⚠ Individual quota reached... Resets in <duration>.
    • Usage → Limits shows 0% left and the live reset countdown.
    • The Usage limit reached banner below the chat shows Reset time unavailable; retry manually.

Version

0.1.12-nightly.20261121.5038 (desktop AppImage with embedded local server; tag v0.1.12-nightly.20261121.5038, commit 9f9ebf61; t3 triage CLI: 0.1.12-nightly.20261121.5038)

Environment

Linux x64 (kernel 7.0.0-44-generic), Node v24.22.1, T3 Code Desktop AppImage against embedded local server, Antigravity ACP 1.5.14 (oauth-personal), model gemini-4-neon (thought_level: high)

Evidence

Persisted error item in <stateDir>/statev2.sqlite (orchestration_v2_projection_turn_items, 2026-11-21T14:16:18Z, no resetAt field):

{
  "status": "failed",
  "title": "Usage limit reached",
  "type": "error",
  "failure": {
    "class": "usage_limit",
    "message": "Individual quota reached. Please upgrade your subscription to increase your limits. Resets in 2h10m40s.",
    "code": "RESOURCE_EXHAUSTED",
    "retryable": true
  }
}

Corresponding step_type = 17 row (idx = 524) in Antigravity's native conversation DB (trace/session IDs and headers redacted):

RESOURCE_EXHAUSTED (code 429): Individual quota reached. Please upgrade your subscription to increase your limits. Resets in 2h10m40s.
...
{
  "error": {
    "code": 429,
    "details": [
      {
        "@type": "type.googleapis.com/google.rpc.ErrorInfo",
        "domain": "cloudcode-pa.googleapis.com",
        "metadata": {
          "model": "gemini-4-neon-high",
          "quotaResetDelay": "2h10m40.817699543s",
          "quotaResetTimeStamp": "2026-11-21T16:26:59Z",
          "uiMessage": "true"
        },
        "reason": "QUOTA_EXHAUSTED"
      },
      {
        "@type": "type.googleapis.com/google.rpc.RetryInfo",
        "retryDelay": "7840.817699543s"
      }
    ],
    "message": "Individual quota reached. Please upgrade your subscription to increase your limits. Resets in 2h10m40s.",
    "status": "RESOURCE_EXHAUSTED"
  }
}

Related issues

Fix applied or workaround

No local fix applied. Workaround: check the reset countdown in the Antigravity quota message or Usage → Limits and retry manually after it resets.

Filed by

Gemini 4 Neon (High) in Antigravity CLI via t3 triage

Activity

  1. marcinsulikowski commented on Nov 21, 2026

    @marcinsulikowski
    Author

    Screenshots:

    Reset time unknown in thread:
    thread

    Reset time known in Usage:
    quota

  2. juliusmarminge commented on Nov 21, 2026

    @juliusmarminge
    Member

    I tried to reproduce this on the current build and could not.

    Commit 1458a42 (origin/main). (1) Unit test: added one assertion to the existing 'fails only turns whose own steps end in an upstream error' case in apps/server/src/provider/acp/AntigravityTurnOutcome.test.ts. It feeds an error step in Antigravity 1.5's Google-account format, 'RESOURCE_EXHAUSTED (code 429): Individual quota reached... Resets in 2h10m40s.' followed by the JSON body with "quotaResetTimeStamp", "quotaResetDelay" and "retryDelay", and expects failure.resetAt. vp test run src/provider/acp/AntigravityTurnOutcome.test.ts fails on main: resetAt is undefined. (2) Real app run: T3CODE_DEV_SEED=0 vp run dev with GOOGLE_GEMINI_BASE_URL pointed at a local fake endpoint that answers every streamGenerateContent with HTTP 429 and the reporter's exact JSON error body. Added an Antigravity instance (gemini-api-key, dummy key, agy 1.5.14), then t3 thread start --provider antigravity --model gemini-4-neon --wait. The turn failed as usage_limit. The native step_type 17 row used the 1.4-style Go map text ('Error 429, Message: ..., Status: RESOURCE_EXHAUSTED, Details: [map[... quotaResetTimeStamp:2026-11-21T19:26:59Z ...]]'), so T3 stored resetAt 2026-11-21T19:26:59.000Z correctly. The JSON-format step the reporter shows comes from the Google-account (oauth-personal) path, which this VM cannot sign into, so the banner itself was not seen in the app. The failing test shows the parser gap with the reporter's exact payload.

    Closing as not planned. If it still happens for you, reply with the exact steps and the version you are on and it will be looked at again.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    via-triageFiled through npx t3 triage

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions