w8123/EnterpriseAgentFramework

面向 OA、ERP、CRM、工单等企业系统快速构建智能体，让 AI 在权限、审计与治理约束下安全调用真实业务能力。Build agents quickly for enterprise systems such as OA, ERP, CRM, and ticketing, enabling AI to safely call real business capabilities under permissions, audit, and governance controls.

What it solves

ReachAI addresses the challenge of integrating AI into existing Java-based enterprise systems (like OA, ERP, CRM, and MES) without sacrificing process determinism, security permissions, audit trails, or cross-system collaboration. It prevents the creation of "AI islands" where AI applications are disconnected from the actual business data, user identities, and existing workflows already embedded in legacy systems.

How it works

ReachAI acts as an AI access and governance layer between the enterprise system and the LLM. It uses a multi-pronged approach:

  • Capability Registration: A Java SDK/Starter allows existing business interfaces and methods to be registered as "Capabilities" without rewriting the services.
  • Deterministic Execution: While AI can draft workflows, the system executes them via GraphSpec, a formal semantic layer that ensures business processes (like expense reports or contract queries) remain predictable and auditable.
  • Identity Bridging: A short-term token mechanism links the platform user, the business system user, and the Agent, ensuring secure cross-system calls.
  • Embedded Interface: A Chat Embed SDK and Page Bridge allow the AI agent to live directly inside the business system's web pages, enabling it to perform "Page Actions" (e.g., filling forms or submitting approvals) in the current context.
  • AI Coding Integration: The framework provides manifests and APIs specifically for AI coding tools (like Cursor or Claude Code) to automate the process of connecting the business system to ReachAI.

Who it’s for

Enterprise developers and architects managing Java-based business systems who want to add AI capabilities while maintaining strict control over business logic and security.

Highlights

  • AI Coding Native: Designed for tools like Cursor and Codex to handle the engineering work of system integration and workflow patching.
  • Graph-Based Determinism: Uses GraphSpec to ensure that AI-generated workflows are converted into stable, executable, and versioned business logic.
  • Low-Intrusion Integration: Uses Spring Boot starters to register existing capabilities with minimal code changes.
  • Context-Aware Page Agents: Enables agents to interact with the actual UI of the business system via Page Bridge.
  • Comprehensive Governance: Includes RunOps, Trace, ACL, and Guard features for full auditability and replay of AI executions.