A modular, CLI-native AI orchestration engine built on the RTX⚡ Core Framework.
Coordinates multiple frontier LLMs through a unified terminal interface —
inspired by Sakana AI's Fugu Ultra multi-agent architecture and Marlin strategic research methodology.
RTXCoreOrchestrate is a CLI-first AI orchestration system that turns your terminal into a multi-agent command center. It routes tasks across frontier LLMs (Claude, GPT, Gemini) using a custom planner → conductor → verifier pipeline — with zero GUI dependencies.
It ships two core components:
| Component | File | Purpose |
|---|---|---|
| RTX Framework | RTXCoreOrchestrate/RTXCoreFramework.md |
Universal AI system prompt — calibrates any LLM |
| RTX Engine | RTXCoreOrchestrate/RTX-Engine/ |
Node.js orchestration core — runs real CLI pipelines |
| Mini Conductor | RTXCoreOrchestrate/mini_fugu_conductor.py |
Lightweight Python Fugu-style task orchestrator |
RTXCoreOrchestrate/ ← Root
├── README.md
└── RTXCoreOrchestrate/ ← Main engine directory
├── RTXCoreFramework.md ← Master AI system prompt (RTX v2.0)
├── mini_fugu_conductor.py ← Python lightweight orchestrator
├── fugu_scratchpads/ ← Session state cache
└── RTX-Engine/ ← Node.js orchestration core
├── config/
│ └── cli-registry.json ← Registered CLI tool definitions
├── core/
│ ├── rtx-conductor.js ← Main task conductor
│ ├── rtx-planner.js ← AAS dynamic step planner
│ ├── rtx-router.js ← Model routing (Standard / Ultra)
│ ├── rtx-mcts.js ← AB-MCTS tree search engine
│ ├── marlin-conductor.js ← Strategic research report generator
│ ├── rtx-verifier.js ← Output quality auditor
│ ├── rtx-safety.js ← Pre-flight security scanner
│ ├── rtx-session.js ← Session state manager
│ ├── rtx-file-manager.js ← Workspace file manager
│ ├── rtx-sanitizer.js ← Input/output sanitizer
│ ├── rtx-cli-bridge.js ← CLI process bridge layer
│ └── adapters/
│ ├── BaseAdapter.js ← Abstract adapter interface
│ ├── ClaudeAdapter.js ← Anthropic Claude CLI adapter
│ └── GhAdapter.js ← GitHub CLI adapter
├── scripts/
│ └── health-check.js ← System health diagnostics
└── workspaces/ ← Auto-generated session outputs
The RTX⚡ Core Framework is a universal system prompt that calibrates any AI agent into a high-performance assistant. Works with Claude, GPT, Gemini, Codex, Cursor, Antigravity, and any LLM that accepts a system prompt.
Core Protocol (RTX):
| Letter | Meaning | Role |
|---|---|---|
| R | Reasoning | Logic verification, problem decomposition, intent understanding |
| T | Thinking | Self-assessment, learning, continuous refinement |
| X | Xtreme | Action-oriented execution, maximum speed and output quality |
How to use:
Copy the full contents of
RTXCoreOrchestrate/RTXCoreFramework.mdand paste it as a System Prompt or Custom Instructions in any AI tool.
On first load, the agent runs the First-Boot Protocol — asking 3 sequential questions (mother tongue, agent name, addressal preference) to permanently configure its behavior.
Inspired by Sakana AI's Fugu Ultra, the engine dispatches subtasks to specialized CLI workers using a multi-stage pipeline:
┌─────────────┐
User Prompt ──►│ rtx-router │ ◄── Selects: Standard / Ultra mode
└──────┬──────┘
│
┌──────▼──────┐
│ rtx-planner │ ◄── AAS: Generates dynamic step plan
└──────┬──────┘
│
┌──────▼────────┐
│ rtx-conductor │ ◄── Dispatches to CLI workers
└──┬────────┬───┘
│ │
┌─────────▼─┐ ┌──▼──────────┐
│ Candidate A│ │ Candidate B │ ◄── Parallel generation
│ (Claude) │ │ (GH/Other) │
└─────────┬──┘ └──┬──────────┘
│ │
┌───▼────────▼───┐
│ rtx-mcts │ ◄── AB-MCTS: Selects best candidate
└───────┬────────┘
│
┌───────▼────────┐
│ rtx-verifier │ ◄── Quality audit + reflection loop
└───────┬────────┘
│
Final Output ✅
| Tool | Version | Install |
|---|---|---|
| Node.js | v18+ | nodejs.org |
| Python | 3.10+ | python.org |
| Claude CLI | Latest | npm install -g @anthropic-ai/claude-code |
| GitHub CLI | Latest | cli.github.com |
# Clone the repository
git clone https://github.com/PsProsen-Dev/RTXCoreOrchestrate.git
cd RTXCoreOrchestrate/RTXCoreOrchestrate/RTX-Engine
# Run system health check
node scripts/health-check.jsnode core/rtx-conductor.js --ultra "Build a FastAPI backend with JWT authentication"node core/marlin-conductor.js "Assess geopolitical impact of Strait of Hormuz blockade"node core/rtx-mcts.js --depth 3 --width 2 "Best architecture for a real-time trading system"cd ../
python mini_fugu_conductor.py "Summarize and compare these two research papers"Marlin-style research reports follow a strict 3-layer analytical structure:
| Layer | Label | Description |
|---|---|---|
| ✅ Layer 1 | Verified Facts | Hard data cross-referenced with citations |
| 🔶 Layer 2 | Working Assumptions | Unverified but plausible inputs |
| 💡 Layer 3 | Analyst Inferences | Reasoned conclusions drawn from above |
Monitoring Horizon Tiers:
| Tier | Window | Focus |
|---|---|---|
| Tier 1 | 0–6 months | Immediate leading indicators |
| Tier 2 | 0–12 months | Medium-term developments |
| Tier 3 | 6–36 months | Long-term structural shifts |
- Pre-flight payload scanning on all shell commands before execution
- Credential filtering — no AWS/GCP/SSH keys leaked to CLI APIs
- Input validation on agent names and all prompt arguments
- License auditing on injected boilerplate dependencies
- Full security policy documented in
RTXCoreFramework.md— Section 14
- Sakana Fugu Technical Report - Multi-Agent System as a Model
- TRINITY Paper - Evolved Model Coordination
- Conductor Paper - RL-Trained Orchestrator
- VentureBeat: Fugu Launch
- AyAutomate: What is Sakana Fugu?
- MarkTechPost: Fugu Orchestration Model
- MindStudio: Multi-Agent AI System
- Paddo.dev: Critical Analysis of Fugu
MIT License — Free to use, modify, and distribute with attribution to the original author.
Built with ⚡ by Prosenjit Paul (aka Prosen)
"Reasoning. Thinking. Xtreme."