[Feature]:Replace deprecated google-gemini-cli provider with antigravity-oauth (Cloud Code Assist 404)

[Primuez]

Problem or Use Case

Google officially sunset the Gemini CLI and replaced it entirely with Antigravity 2.0 and the agy CLI. As a result, the backend endpoint (cloudcode-pa.googleapis.com) that Hermes relies on for the google-gemini-cli provider is returning a hard HTTP 404: Requested entity was not found.

This completely breaks the ability to use a Google AI Pro subscription for agent model routing via OAuth.

Steps to Reproduce:

Authenticate using hermes auth add google-gemini-cli --type oauth (or inject a valid refresh token).

Run any prompt via the CLI: hermes -m google-gemini-cli:gemini-3.5-flash -z "Test"

Check logs. Hermes swallows the zero-shot output, but debug logs reveal the endpoint is dead: API call failed: HTTP 404: Code Assist 404... is not available at cloudcode-pa.googleapis.com

Why this is critical:
For users building multi-agent orchestration systems on weak VPS instances—specifically relying on Hermes to manage standard webhook tools for SaaS backends like Primuez Policy Decoder—the Pro subscription OAuth route is the only way to avoid massive API costs while maintaining high token limits. Using provider: auto or fallback API chains isn't a sustainable replacement for the native Pro quota.

Proposed Solution

Requested Solution:

Deprecate the google-gemini-cli provider string.

Introduce a new antigravity-oauth provider.

Update the internal Google Code Assist adapter to route through the new Antigravity endpoints rather than the retired cloudcode-pa URLs.

Alternatives Considered

No response

Feature Type

Configuration option

Scope

None

Contribution

• I'd like to implement this myself and submit a PR

Debug Report (optional)

[kzeokytjvwhnrtl13otf-bit]

This is now a live outage, not a future concern.

The Gemini CLI Code Assist endpoint was sunset on June 18 (2 days ago). Every consumer-tier user of google-gemini-cli is currently broken:

• gemini-3.5-flash → HTTP 404 (hard fail, every request)
• gemini-3.1-pro-preview → HTTP 400 or 429 (intermittent, function call mismatch errors)
• gemini-3-pro-preview → Same as above

The CLI path still works for some models on some requests, but the desktop GUI is completely unusable with Gemini — every prompt errors out.

There are already two PRs open for this since May 20 — a full month before the sunset:

• feat(antigravity): add Antigravity CLI OAuth provider #29385 (feat/antigravity-cli-provider) — OPEN, unreviewed, last updated May 23
• feat(cli): add native Antigravity OAuth provider #29474 (feat/antigravity-native-oauth) — OPEN, merge conflicts, last updated June 12

Both have been sitting untouched for weeks. Meanwhile, users who paid for Google AI Pro subscriptions specifically to use Gemini models through Hermes have no working path. The only options right now are:

1. Switch to Xiaomi MiMo (works, but not what we subscribed for)
2. Use third-party API proxies (fragile, against ToS, defeats the purpose of a Pro subscription)
3. Stop using Hermes for Gemini entirely and go back to Antigravity CLI directly

Please prioritize merging one of the existing PRs or implementing the Antigravity OAuth provider. The window for "this is a nice-to-have" closed on June 18. For consumer-tier subscribers, Hermes's Google integration is currently dead.

P1 or P2 severity — this is a regression that breaks a core provider for paying users.

[Primuez]

Bro I found a solution called. Open route not open router provider but open route which connects with my agy