Agent chat completely broken: "neither PlanModel nor RequestedModel specified" on every message (daily-cloudcode-pa backend)

Every message I send in the Agent panel fails instantly with “Failed to send.” This happens in brand-new conversations and existing ones, regardless of which model I pick from the model dropdown (tried Gemini 3.6 Flash, 3.5 Flash, 3.1 Pro).

Checked the Output panel’s “Antigravity” log channel and found the actual server-side error:

neither PlanModel nor RequestedModel specified. You must specify a valid model.

The app is connecting to https://daily-cloudcode-pa.googleapis.com (visible in the “cloudcode” output channel on startup).

What I’ve already tried, none of which fixed it:

  • Fully quit and relaunched the app
  • Re-selected the model explicitly from the dropdown before sending
  • Signed out of my Google account, closed the app, reopened, and signed back in
  • Submitted an in-app bug report (Settings > Provide Feedback) with server logs attached

Version: 2.3.1 (July 16, 2026 release, per the changelog) OS: Windows Account: signed in via Google Auth

This looks like a client/backend mismatch on the daily channel rather than anything account- or machine-specific, since it reproduces identically across a full re-auth and reinstall-equivalent relaunch. Would appreciate a look — right now the Agent panel is completely unusable.

Hi @Julian_Elias,

Thank you for reporting this and providing the detailed logs.

To resolve this, could you please follow the below steps:

1. **Update Antigravity**: Please update to the latest build (2.8+ or newer), where model resolution and session schemas have been updated.

2. **Reset Onboarding State**:

- Open the Command Palette (`Ctrl+Shift+P`).

- Run **`Antigravity: Reset onboarding`**.

- Complete the setup and sign-in flow again to refresh your cached session configurations.

3. **Start a Fresh Workspace Session**: If an older conversation thread was corrupted with the empty model configuration, opening a new chat session after resetting will ensure clean cascade parameters are used.

Please let us know if the issue persists after updating.