SlimeTree-RLM Prompt Gateway

Stop wasted shots from your internal ChatGPT

Sign in with a Meta account → the in-browser RLM classifies (D / µ / R) → only R is delegated to the Claude API = API cost is reduced structurally. All processing runs in the browser; no JAVATEL server hop.

1. Sign in / Trial sign up

RECOMMENDED · Fastest route

Instant trial with a Meta account

Use your Facebook / Instagram sign-in directly. Permissions are public_profile + email only; one email confirmation auto-starts a 30-day SlimeTree-RLM trial.

— or another method —

Real SlimeTree-RLM trial — ID + Password authentication, with a SHA-256 audit chain + ed25519 license issued by the JAVATEL license server.
New users: enter email + password → "Sign up" → click YES in the confirmation email to auto-start the 30-day free trial.
Existing users: use "Sign in" in the same form.

Email address

Password (8 chars min)

Extra fields required for sign-up (Name / Country / Individual or Corporate)

Name

Country (ISO 2-letter)

Account type

Company name (required for Corporate only)

I agree to the Terms of Service and Privacy Policy

DEMO —

SLIMETREE-RLM SUBSCRIPTION

Checking...

2. LLM API key (used on R verdict)

The LLM is only called on R verdicts (LLM delegation). Keys are stored only in this browser's localStorage; nothing is ever sent to JAVATEL servers (browser-direct call).

Active provider: Gemini (Google) Everyone OK Claude (Anthropic) token billing OpenAI token billing

LLM-neutral: free choice of Gemini / Claude / OpenAI / Grok — Gemini = free tier available; Claude / OpenAI = each provider's token billing. Fees go directly to your LLM provider; JAVATEL does not mediate.

GEMINI (Google)

CLAUDE (Anthropic) · token billing

OPENAI · token billing

Note: Even with no key set, D / µ verdicts still work. R-verdict messages will just display a key missing message. Billing goes directly to your provider account (Google / Anthropic / OpenAI); JAVATEL does not mediate.

2.5 Operating mode NEW: Pattern B (cross-vendor)

Single LLM — call only the single active provider above Cost-tier escalation (B) — cheap attempt → RLM re-judges → premium only when insufficient

3. Prompt input

Examples: What is the elevation of Mt. Fuji? → D / あ → µ / What are the 5 most historically influential inventions? → R

0Total queries

0D (instant)

0µ (suppressed)

0R (LLM delegation)

0R cache hits

0Actual LLM calls

0cheap (B)

0escalated (B)

—esc rate (B)

SlimeTree-RLM pipeline (real-time)

☀

Input

▸

µ scan

▸

D lookup

▸

R decide

▸

∶

WAL chain

▸

☀

Claude API

▸

➤

Delivery

D = instant (no LLM call) µ = suppressed (no LLM call) R = LLM delegation (free choice of Gemini / Claude / OpenAI)

Type a prompt and the RLM verdict + reply will appear here.

4. Audit / WAL operations

Show audit / export result

(nothing has run yet)

5. Why this gateway is "cost reduction + safety"

D verdict (structurally certain facts): respond instantly without calling Claude → 0 token cost
µ verdict (insufficient grounds / suppression-worthy): respond with a suppression message, no Claude call → 0 token cost + 0 prompt-injection surface
R verdict (genuinely needs an LLM): delegate to Claude → billed only for what is needed
WAL audit chain: every query is recorded in a SHA-256 chain → consistent audit even air-gapped
Browser-only: zero server-side ops cost; Meta sign-in only certifies the user is real

For the same prompt, R is cached (sessionStorage) and billed only once. In real production you would persist the exportWAL() result to S3 and instantly reproduce the past chain state via rollback().