NEW · 2026-05-31 MIT 6-pattern showcase Gemini-only (demo simplicity)

Multi-Agent Demo — 6-pattern showcase

Run all 6 patterns of using SlimeTree-RLM as a multi-agent orchestrator (A cross-validation / B cost-tier / C specialist routing / D orchestrator-worker / E debate / F voting) side by side on one page. Same prompt, 6 ways, easy to compare.

One Gemini API key covers all 6 patterns (issue a personal free key at aistudio.google.com/apikey; Free tier 15 RPM).

0. Gemini API key (one key, all 6 patterns)

The key is stored only in this browser's localStorage; it is never sent to JAVATEL servers.

1. Shared prompt (each pattern processes the same prompt = comparable)

Samples:

A Cross-validation

Two LLM calls (low temp + high temp) → compare outputs; agreement = adopt / disagreement = flag for human review. For high-precision requirements (finance / medical / legal).

(not yet run)

B Cost-tier escalation

Try on cheap (Flash) → RLM judges quality → escalate only insufficient items to premium (Pro). 50-70% layered cut on top of the existing 73% cut.

(not yet run)

C Specialist routing

Classify the prompt by keyword into a domain (code / legal / medical / general) → apply a specialist system_prompt and call Gemini. Enterprise multi-domain support.

(not yet run)

D orchestrator-worker

The orchestrator (Flash) splits the prompt into 2-3 sub-questions → parallel workers (Flash) handle each sub-question → merge. Parallelizes large tasks.

(not yet run)

E debate / critique loop

A proposes → B critiques → A revises, up to 2 iterations. RLM detects convergence / divergence and caps the loop. For code review / document drafting.

(not yet run)

F Voting / consensus

Send the same prompt to Flash three times at different temperatures (0.2 / 0.5 / 0.9) in parallel → majority vote on response fingerprints. Agreement rate = confidence score.

(not yet run)

★ Run all 6 patterns at once

Send the shared prompt through all 6 patterns simultaneously and compare results side by side. Approximately 12-15 API calls (sits within Free tier 15 RPM; use carefully).

2. Global API call log (across patterns)

(not yet run)

3. Why RLM works as an orchestrator

  • Drops in without modification ― subdivides the R of the existing D/μ/R verdict; the same patch applies across all 18 routes.
  • Audit doesn't break ― the SHA-256 WAL chain records every agent call; instant regulatory compliance.
  • Preserves Platform-native ― Meta = cross-vendor freedom, X = in-Grok, Google = in-Gemini, each constraint honored.
  • Natural fallback ― if premium fails, automatically falls back to the cheap response; no extra wiring required.

Platform Integrations Hub → Layered-cut numbers → Meta Gateway (working B impl) →