labsai/EDDI

What it solves

EDDI (Enhanced Dialog Driven Interface) is a production-grade middleware designed to govern non-deterministic AI agents in a deterministic way. It solves the challenge of moving multi-agent systems from prototypes to production by providing a config-driven orchestration engine that emphasizes security, compliance (GDPR, HIPAA, EU AI Act), and auditability without requiring custom application code for every agent.

How it works

Built with Java 25 and Quarkus, EDDI acts as an orchestration layer between users, AI agents, and business systems. It uses versioned JSON configurations to define agent behavior, allowing updates without redeployment. The system coordinates agents through intelligent routing, persistent memory, and API orchestration. It supports a wide array of LLM providers (cloud, enterprise, and self-hosted) and implements open standards like the Model Context Protocol (MCP) and Agent-to-Agent (A2A) protocol for interoperability.

Who it’s for

It is designed for developers and enterprises needing a secure, scalable, and compliant multi-agent framework for conversational AI, particularly those operating in regulated industries or those who prefer a Java-based infrastructure over Python/Node libraries.

Highlights

• Multi-Agent Orchestration: Supports intelligent routing, nested groups, and six built-in discussion styles (e.g., Peer Review, Devil's Advocate) for multi-agent debates.
• Production Governance: Features an HMAC-SHA256 immutable ledger for audit trails, envelope-encrypted secrets vault, and built-in tenant quotas.
• Comprehensive Memory: Implements persistent user memory with "dream consolidation" for background maintenance and rolling summaries for long-term context.
• RAG & Tooling: Integrated support for 7 embedding providers, 6 vector stores, and a suite of built-in tools including web search, PDF readers, and sandboxed calculators.
• Interoperability: Native support for MCP, A2A, OpenAPI 3.1, and OAuth 2.0/OIDC via Keycloak.