OrchestrAI: An AI Agent Orchestration Platform Based on Clean Architecture

OrchestrAI is an open-source AI agent orchestration platform developed and maintained by Kaewdisorn (GitHub link: https://github.com/Kaewdisorn/OrchestrAI, released on 2026-05-25). It is built using Clean Architecture, CQRS, and Hexagonal Design patterns, supporting tool execution, multi-layer memory systems, and scalable multi-agent workflows, implemented with TypeScript and pnpm workspaces. Its core goal is to address engineering challenges in AI system collaboration, maintainability, and scalability.

As the capabilities of large language models evolve, a single model can already solve complex problems, but how to organize them into collaborative, scalable, and maintainable systems has become a new challenge. OrchestrAI aims to strike a balance between the flexibility of AI applications and the maintainability of software systems by applying mature software engineering architecture patterns (such as Clean Architecture, CQRS, and Hexagonal Architecture).

Clean Architecture

Emphasizes separation of concerns and cohesive dependencies:

• Domain Layer: Defines core concepts like agents and tasks, with no external dependencies;
• Application Layer: Orchestrates use cases and business processes;
• Infrastructure Layer: Handles external dependencies like LLM clients and databases;
• Interface Layer: Provides APIs and interactive interfaces.

Hexagonal Architecture

Enhances testability and replaceability through ports (interface contracts) and adapters (concrete implementations), such as OpenAI adapters and PostgreSQL memory adapters.

CQRS

Separates write operations (command side: agent creation, task submission, etc., ensuring transaction consistency) from read operations (query side: status query, history retrieval, etc., which can be optimized independently).

Tool Execution System

Supports synchronous/asynchronous calls, tool validation and secure sandboxing, pluggable tool discovery, typed result processing, and error recovery.

Memory System

Includes working memory (session short-term context), short-term memory (recent cross-session history), long-term memory (refined knowledge with vector retrieval support), and shared memory (multi-agent collaboration).

Multi-agent Workflow

Supports hierarchical collaboration (main agent coordinating sub-agents), peer-to-peer collaboration (equal negotiation), pipeline collaboration (process chaining), and dynamic collaboration (runtime team formation), along with production-grade features like state persistence and checkpoint resume.

Full-stack TypeScript

100% TypeScript usage, bringing advantages like type safety, IDE support, and Node.js ecosystem integration.

pnpm Workspace

Uses monorepo management:

• apps/api: Main API application;
• k8s/postgres: K8s deployment configuration and database resources;
• Shared packages with independent version management and dependency deduplication optimization.

Turborepo

Enables intelligent task orchestration, remote caching, and incremental builds via turbo.json to speed up CI/CD.

Kubernetes Ready

Provides K8s configuration and PostgreSQL persistence solutions, supporting cloud-native deployment.

Documentation-Driven Development

Includes planning documents like plan.md and slice1.md to maintain team alignment and guide AI-assisted development.

OrchestrAI is suitable for:

1. Complex Task Automation: Orchestrating multi-step, multi-tool collaborative tasks (e.g., data analysis report generation);
2. Intelligent Team Collaboration: Building professional agent virtual teams to collaboratively complete comprehensive tasks;
3. Scalable AI Services: Smoothly scaling from prototype to production-grade services based on Clean Architecture;
4. Enterprise AI Integration: Flexibly integrating internal systems (authentication, data sources, etc.) via Hexagonal Architecture.

OrchestrAI is currently in the early stage (0 stars, 0 forks). Early participants can deeply influence the project direction, and the codebase is concise and easy to contribute to.

It represents the evolutionary trend of AI applications from "prompt engineering" to "system engineering", providing an architectural reference for production-grade agent applications. As multi-agent systems become widespread, such orchestration platforms will become an important part of AI infrastructure, worthy of attention and evaluation by technical decision-makers.

It is recommended that developers interested in agent orchestration and AI system engineering follow and participate in this project.