Powered by OpenAirymax
The seminal fourth "Operating System Philosophy" in human computing history.
Language: English | 简体中文
| User Type | Product | Version | Download |
|---|---|---|---|
| Pro Users | Docker | v0.1.0 |
📦 Get Docker |
| Personal Users | Desktop | v0.1.0 |
🖥️ Get Desktop |
AgentRT is an intelligent agent runtime platform that provides comprehensive OS-level support for orchestrating agent teams.
1.1 Project Preview
Personal Client Preview
⚡️
Foundational Theory Multibody Cybernetic Intelligent System
- Pure Kernel: Kernel only provides atomic mechanisms, pure and efficient
- Cognitive Loop: Cognition, Planning, Action
- Memory Stratification: L1 Raw Layer → L2 Feature Layer → L3 Structure Layer → L4 Pattern Layer (OSS: L1+L2 built-in, PRO: L1-L4 via MemoryRovol)
- Inherent Security: Sandbox isolation, permission arbitration, input sanitization, audit trail
- Token Efficiency: Saves approximately 500% tokens compared to traditional frameworks
- Comprehensive SDKs: Native support for Go, Python, Rust, TypeScript
3.1 Team Drive
- Precisely coordinates multi-Agent collaboration
- Efficiently completes complex task orchestration and resource scheduling
3.2 Autonomous Evolution
- Possesses self-evolution capability
- Dynamically adjusts strategies
- Continuously optimizes execution effectiveness
✨
Brand New Architecture · Inherent Security · Intelligence Emergence
4.1 Architecture Design Complete architecture from kernel to application:
⬇️ Ecosystem Layer (ecosystem) — Apps/Config/Prompts/Hooks/Skills
⇅ Service Layer (daemon) — 12 daemon services
⇅ Protocol Layer (protocols) — 5-layer unified protocol stack
⇅ Gateway Layer (gateway) — HTTP/WS/Stdio → JSON-RPC 2.0
⇅ Storage Layer (heapstore) — Runtime data persistence
⇅ Security Layer (cupolas) — 4-layer inherent security
⇅ Kernel Layer (atoms) — 7 atomic modules
⇅ Support Layer (commons) — Unified foundation library
⬆️ SDK Layer (sdk) — Python/Go/Rust/TypeScript
4.2 Design Principles Built upon ARCHITECTURAL_PRINCIPLES:
- System Perspective: Real-time response <10ms Feedback loops, layered decomposition Holistic design, emergence management
- Kernel Perspective: Kernel ~6K LOC (full repo ~478K LOC) Minimalist kernel, contractual interfaces Service isolation, pluggable strategies
- Cognitive Perspective: Token savings 80% Dual-system synergy, incremental evolution Memory stratification, forgetting mechanism
- Engineering Perspective: Test coverage >95% Security built-in, observability Resource determinism, cross-platform consistency
- Design Aesthetics: API <50 per module Simplicity first, extreme attention to detail Human-centric, perfectionism
5.1 Environment Requirements
- Operating System: Ubuntu 22.04+ / macOS 13+ / Windows 11 (WSL2)
- Compiler: GCC 11+ / Clang 14+ (C11/C++17)
- Build Tools: CMake 3.20+, Ninja
- Python: 3.10+ (Required for OpenLab)
5.2 Installation & Build
# 1. Clone repository
git clone https://atomgit.com/openairymax/agentos.git && cd agentos
# 2. Install dependencies (Ubuntu)
sudo apt install -y build-essential cmake gcc g++ libssl-dev libsqlite3-dev ninja-build
# 3. Build kernel (out-of-source build required by BAN-33)
mkdir /tmp/AgentRT-build && cd /tmp/AgentRT-build
cmake /path/to/AgentRT -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTS=ON
cmake --build . --parallel $(nproc)
# 4. Run tests
ctest --output-on-failure5.3 Docker Quick Start
# Build image
docker build -f deploy/docker/Dockerfile -t agentrt:latest .
# Run container
docker run -d --name agentrt -p 8080:8080 -v ./config:/app/config agentrt:latest
5.4 Usage Methods
| Language | Usage Method |
|---|---|
| C/C++ | Develop via syscalls.h system call interface |
| Python | Install via pip install agentos then directly import |
| Go | Use import "github.com/spharx/agentos/sdk/go" |
| Rust | Use use agentrt_toolkit::prelude::*; |
| TypeScript | Install via npm install @spharx/agentrt-toolkit then directly import |
5.5 Reading Navigation
| Document | Core Content |
|---|---|
| 📘 Architectural Principles | Five-dimensional orthogonal system, 24 core principles |
| 🚀 Quick Start | 5-minute getting-started guide |
| ⚙️ Build Guide | Detailed build steps and options |
| 🧪 Testing Guide | Unit/Integration/Contract testing |
| 🐳 Deployment Guide | Docker/Kubernetes deployment |
5.6 Common Questions
👉 Q1: What is the difference between AgentRT and traditional AI Agent frameworks?
AgentRT is an operating system-level product, not a single framework:
| Dimension | AgentRT | Traditional Frameworks |
|---|---|---|
| Positioning | Multi-agent collaboration OS | Single agent |
| Architecture | Microkernel + strict layering | Loosely coupled modules |
| Security | Four-layer inherent security | Application-level protection |
| Memory | Four-layer stratification system | Vector database |
| Token Efficiency | Saves approximately 500% | No optimization |
👉 Q2: Which application scenarios is it suitable for?
✅ Especially suitable
- 🎯 Complex multi-step task orchestration
- 🧠 Long-term memory and knowledge accumulation needs
- 🔒 High-security enterprise applications
- 💾 Resource-constrained embedded scenarios (atomslite)
- 🌐 Multi-language development teams
❌ Not suitable
- 🚫 Simple single-call tasks (using a sledgehammer to crack a nut)
👉 Q3: How is security guaranteed?
Security built-in design, four-layer protection
| Protection Layer | Implementation Method |
|---|---|
| Virtual Workspace | Process/Container/WASM sandbox isolation |
| Permission Arbitration | RBAC + YAML rule engine |
| Input Sanitization | Regex filtering + Type checking |
| Audit Trail | Full-chain tamper-proof logging |
👉 Q4: What prerequisite knowledge is needed for learning?
| Role | Prerequisite Knowledge | Time to Get Started |
|---|---|---|
| Application Developer | Python/Go basics | 1-2 days |
| System Developer | C/C++, OS fundamentals | 1-2 weeks |
| Architect | Microkernel, distributed systems | 1 month |
Recommended path: Quick Start → Architectural Principles → CoreLoopThree
We are walking into the future: "Intelligence emergence, and nothing less, is the ultimate sublimation of AI".
The power of belief
☀️
This is not humanity's sunset, but the dawn of a new world
Believe: The spirit of open source can maximize the wisdom of the group; Collaboration will propel humanity to new heights.
Witness: Every day of our work is part of history; It will surely be engraved on the monument of human civilization's development.
Contribution: Whether you are an experienced developer or just starting out:
Discover: Report bugs, help us improve quality
Ideas: New feature suggestions, make the project stronger
Share: Improve documentation, help more people understand AgentRT
Coding: Submit PRs, jointly create history
🔥
A faint light cannot illuminate the entire path, yet it guides our direction forward
6.1 Contribution Process: See Contributing Guide
Fork Project → Create Branch → Develop & Test → Submit PR → Code Review → Merge to Main
Main Platforms: AtomGit (Recommended) · Gitee · GitHub
6.2 Contributors: See AUTHORS.md for the list of contributors.
This project is dual-licensed under AGPL v3 + Apache 2.0 (SPDX: AGPL-3.0-or-later OR Apache-2.0). You may choose either license at your option. See LICENSE file for details. Copyright (c) 2025-2026 SPHARX Ltd.
"From data intelligence emerges."
From data, to intelligence.
AtomGit · Gitee · GitHub · Official Website
© 2026 SPHARX Ltd. All Rights Reserved.