labsai/EDDI

Config-driven engine that turns JSON into production-grade AI agents. Multi-agent orchestration, 12+ LLM providers, MCP/A2A protocols, RAG, persistent memory, and enterprise compliance (EU AI Act, GDPR, HIPAA). Built on Quarkus.

What it solves

EDDI (Enhanced Dialog Driven Interface) is a production-grade middleware designed to govern non-deterministic AI agents within a deterministic engine. It solves the problem of moving multi-agent frameworks from prototypes (often built as Python/Node libraries) to governed, secure, and compliant production environments by providing a config-driven orchestration layer that avoids dynamic code execution.

How it works

EDDI acts as an orchestration layer built with Java 25 and Quarkus. It coordinates users, AI agents, and business systems using versioned JSON configurations rather than embedded application code. It supports a wide array of LLM providers, vector stores, and embedding providers. The system implements open standards like the Model Context Protocol (MCP) and Agent-to-Agent (A2A) protocol to ensure interoperability. It manages complex agent interactions through intelligent routing, persistent memory, and a structured "group conversation" system for multi-agent collaboration.

Who it’s for

It is designed for organizations requiring production-ready conversational AI that must adhere to strict security, compliance (GDPR, HIPAA, EU AI Act), and audit requirements, as well as developers who want to deploy multi-agent systems without writing extensive custom orchestration code.

Highlights

  • Multi-Agent Orchestration: Supports complex interaction styles like Peer Review, Devil's Advocate, and Negotiation, including group voting and shared artifacts.
  • Config-Driven: Agent behavior is updated via JSON configurations, allowing for changes without redeploying code.
  • Enterprise-Grade Infrastructure: Features an envelope-encrypted secrets vault, HMAC-SHA256 immutable audit ledgers, and native Kubernetes/OpenShift support.
  • Broad Interoperability: Native support for MCP, A2A, OpenAPI 3.1, and OAuth 2.0/OIDC.
  • Advanced Memory Management: Includes persistent user memory, "dream consolidation" for background memory maintenance, and token-aware context windowing.
  • Flexible LLM Support: Integrates with 12 providers across cloud, enterprise, and self-hosted (Ollama, Jlama) options, including multi-model cascading.
  • Integrated RAG: Supports 7 embedding providers and 6 vector stores, including a REST ingestion API.

Related

  • Project
    landingbj/LinkMind

    LinkMind is an enterprise-grade multimodal AI middleware that provides a stable layer between business systems, model providers, and agent runtimes to simplify integration and governance.

  • Project
    soumatheusgomes/vibe-coding-toolkit

    A documented workflow and set of tools (plugins, prompts, lint rules, sub‑agent orchestration, memory handling) that let developers use Claude Code / Codex as a disciplined coding teammate. Includes a “Superpowers” discipline, parallel sub‑agents, quality‑gate ESLint config, and a full onboarding Playbook.

  • Project
    zts212653/clowder-ai

    A self-hosted workspace for coordinating teams of AI agents from different model families, providing persistent identity, shared memory, and cross-model review.

  • Project
    labring/aiproxy

    A production-ready AI gateway that provides a unified, OpenAI-compatible interface for routing, monitoring, and multi-tenant management across multiple AI providers.

  • Project
    ag2ai/ag2

    An open-source AgentOS framework for building AI agents and orchestrating their cooperation through a hub-and-channel network system.