PublicMITGemini-only

Google Prompt Gateway × SlimeTree-RLM

▸ Operate Google safely through an internal AI gateway

Google account sign-in → prompt submission → in-browser RLM emits D/µ/R verdict → only R is delegated to Gemini. Single-account ops under the Google-native LLM principle.

1. Google API authentication (OAuth 2.0)

At console.cloud.google.com create an OAuth client → obtain a User Access Token. Scope: https://www.googleapis.com/auth/userinfo.email

2. Gemini (Google AI) API key

At aistudio.google.com/apikey you can issue a personal free key (15 RPM free; overflow is token-billed).

Platform-native LLM: Google Workspace processing is Gemini-only (managed under the same Google account).

2.5 Operation mode ― ★ NEW: B pattern

Single LLM ― call only the one model selected above Cost-tier escalation (B) ― cheap attempt → RLM re-judge → only the shortfall goes to premium

3. Prompt submission (D=instant / µ=suppressed / R=Gemini)

D: 0

µ: 0

R: 0

Gemini calls: 0

cheap: 0

escalated: 0

escalation rate: ―

Submit a prompt for RLM verdict + Gemini delegation. In B mode the 2-stage pipeline is visualized.

4. API call log

(not yet run)

5. Benefits

Single-account ops: Google + Gemini unified management, no separate billing
D instant: 0-token immediate reply
µ suppression: speculative / sensitive prompts mechanically suppressed
R Gemini: only what truly needs it
WAL chain: SHA-256 audit (in cost-tier mode each cheap + premium step is recorded)
★ B-mode additional cut: cheap (Flash) answers 85% instantly → only the remaining 15% escalates to premium (Pro) → see service page #multi-agent-saving (with premium=Pro config the R portion is further -77.9% / 94.0% combined)