Zing 论坛

正文

HPS AI Hub:面向多领域专业工作的AI操作系统蓝图

一个将AI代理工作流系统化的开源参考架构,通过多模型路由、100+专业代理和跨会话持久记忆,为独立开发者管理8个专业领域提供统一解决方案。

AIagentworkflowSDDmulti-modelroutingAstroTailwindTypeScriptopen source
发布时间 2026/04/14 06:45最近活动 2026/04/14 06:52预计阅读 4 分钟
HPS AI Hub:面向多领域专业工作的AI操作系统蓝图
1

章节 01

HPS AI Hub: A Blueprint for AI Operating System for Multi-Domain Professional Work

HPS AI Hub is an open-source reference architecture designed as an 'AI operating system' to help independent developers/small teams manage complex workflows across 8 professional domains. Its core features include multi-model routing, 100+ professional agents, cross-session persistent memory, and a 'document-first' approach (Spec-Driven Development) to address AI engineering complexity and technical debt.

2

章节 02

Architecture Vision: Unifying 8 Domains with DevOps-Inspired Workflow

The project aims to unify workflows for 8 domains covering full-stack needs: Build (code/architecture), Operate (ops/monitoring), Grow (marketing/business), Secure (audit/compliance), etc. This extends DevOps principles to broader business areas. It uses multi-model routing, 100+ agents, and cross-session memory to manage these domains. Notably, it adopts a €68/month subscription model, balancing technical experiment and commercial sustainability.

3

章节 03

Spec-Driven Development (SDD): 7-Stage Pipeline & Multi-Model Collaboration

HPS AI Hub's key methodology is SDD, a 7-stage pipeline: explore → propose → [spec ∥ design] → tasks → apply → verify → archive. Each stage uses specific models: Opus for architecture decisions, Sonnet for code analysis/spec writing/task splitting/verification, Haiku for archiving. The 'apply' stage routes tasks to specialized models (Gemini 3.1 Pro for visual engineering, GPT-5.4 for analysis, Codex for execution, etc.) via Atlas.

4

章节 04

Technical Stack & Core Features: Persistent Memory & Model Routing

Tech stack: Astro 6.1.4 (static + React islands), Tailwind CSS4 (declarative @theme tokens), TypeScript, Node.js ≥22.12.0, Vercel deployment. Key features: 1) Persistent memory via Engram (SQLite+FTS5) for cross-session context retention; 2) OmO 3-layer model routing (top: Prometheus/Metis/Momus for decision, middle: Atlas for coordination, bottom: category workers for execution) to optimize model usage (including free models like Qwen 3 Coder 480B FREE).

5

章节 05

Open Source Ecosystem & Business Model

HPS AI Hub integrates with open-source projects: gentle-ai (Go-based configurer), Engram (memory system), opencode.ai (runtime), oh-my-openagent (OmO orchestration). It uses MIT license and is maintained by HPS Proptech & Solutions B.V. (Amsterdam/Seville). The business model is €68/month subscription for indie devs/small teams.

6

章节 06

Conclusion: Methodological Value for AI Engineering

HPS AI Hub's value lies in its方法论 exploration for AI engineering, addressing the 'tool chasing' trap. Its document-first approach, SDD pipeline, and multi-model routing provide a systematic way to integrate AI into workflows. Though in early stages (v0.1.0, design system complete, content migration ongoing), it offers a reference for teams looking to systemize AI agent workflows.