Production-grade, config-driven multi-agent orchestration middleware for conversational AI, featuring group debates, intelligent routing, MCP/A2A protocol interoperability, and enterprise security compliance.
E.D.D.I (Enhanced Dialog Driven Interface) is built on the philosophy "The engine is strict so the AI can be creative," delivering comprehensive multi-agent orchestration. The system includes 5 built-in group discussion modes (Round Table, Peer Review, Devil's Advocate, Delphi, Debate) and supports nested group orchestration for complex topologies like tournament brackets and red-team vs. blue-team scenarios. Agent creation is achievable through natural conversation with the built-in Agent Father meta-agent, and all agent behaviors are driven by versioned JSON configurations without code.
The system supports 12+ LLM providers (OpenAI, Anthropic Claude, Google Gemini, Mistral AI, Azure OpenAI, Amazon Bedrock, Oracle GenAI, Google Vertex AI, Ollama, Jlama, Hugging Face, and any OpenAI-compatible endpoint) with Smart Model Cascading for cost optimization—attempting cheaper models first and automatically escalating when confidence is low, complemented by Per-Conversation Budgets and Tenant Cost Ceilings for fine-grained cost control.
For standard protocols, E.D.D.I simultaneously implements MCP (42-tool Server + Client, controllable from Claude Desktop and Cursor), A2A (cross-platform agent communication and skill discovery), and OpenAPI 3.1 (paste a spec to generate a full API-calling agent). Memory management covers persistent user memory (global/agent/group visibility scopes), Dream Consolidation background maintenance, Token-Aware Windowing, and Rolling Summary incremental summarization. RAG capabilities support 7 embedding providers and 5 vector stores, plus zero-infrastructure httpCall RAG.
Security features include an envelope-encrypted Secrets Vault (PBKDF2 + AES-256), SSRF protection, sandboxed expression evaluation, Ed25519 agent signing, and HMAC-SHA256 immutable audit ledgers, with built-in compliance infrastructure for GDPR, HIPAA, and EU AI Act. The runtime is based on Java 25 Virtual Threads and Quarkus 3.35.2, offering one-click Docker installation, multi-overlay Docker Compose deployment, Helm Charts, K8s manifests, and a built-in 45-panel Grafana monitoring dashboard.
Note: The official website (eddi.chat) and documentation site (docs.eddi.chat) are currently unreachable (SSL certificate mismatch / DNS resolution failure). The organizational background of labsai is not explicitly stated in the repository, and specific implementation details of certain compliance claims (e.g., EU AI Act) remain to be confirmed.