Launched this week
EDDI v6
Config-driven multi-agent AI. Compliance-ready, open source
5 followers
Config-driven multi-agent AI. Compliance-ready, open source
5 followers
EDDI v6 is a config-driven multi-agent AI engine. Define agent behavior in JSON, not code. One-command install, running in under 2 minutes. 12 LLM providers (OpenAI to Ollama). MCP server (42 tools) + client. A2A protocol. Persistent memory. Group debates with 5 styles. Model cascading with cost caps. GDPR, HIPAA, EU AI Act compliance built in. Encrypted vault, SSRF protection, zero code execution. Single Docker image. Open source, Apache 2.0.
Why version 6 was built
AI agents outgrew prototyping. They're running in customer support, compliance workflows, healthcare, internal ops. Places where "it mostly works" isn't good enough. But the tooling hasn't caught up. Most frameworks still assume someone will wire everything together in Python, deploy it by hand, and hope the LLM behaves in production.
EDDI v6 was built around one question: what does it take to make multi-agent AI safe, governable, and accessible without requiring a platform team to run it?
Three things:
🚀 Accessibility. Previous EDDI versions were powerful but required deep knowledge to set up. v6 ships with a one-command installer. Full stack (engine + database + starter agent) running in under 2 minutes. Agents can be created through a visual UI wizard, programmatically via REST/MCP, or through natural conversation with the built-in Agent Father meta-agent. No YAML, no boilerplate.
🔗 Interoperability. AI agents shouldn't live in silos. v6 implements MCP (Model Context Protocol) as both server (42 tools) and client, plus A2A (Google's Agent-to-Agent protocol) for cross-platform skill discovery. EDDI agents are controllable from Claude Desktop, Cursor, or any MCP-compatible tool, and can discover and talk to agents on other platforms.
🛡️ Production governance. The part that's usually left as an exercise for the reader. v6 ships with GDPR cascading erasure, HIPAA deployment guidance, EU AI Act audit trails (HMAC-SHA256 immutable ledger), an envelope-encrypted secrets vault (AES-256), SSRF protection on every tool, and per-tenant cost budgets. Not just documentation. Enforced by the architecture. Zero dynamic code execution, by design.
Beyond these three pillars, v6 is the most developer-friendly EDDI release ever:
🤖 12 LLM providers with a unified config model. Switch from OpenAI to Ollama by changing one field 🧠 Persistent memory across sessions with dream consolidation, rolling summaries, and token-aware windowing 🗣️ Multi-agent group conversations with 5 debate styles (Round Table, Devil's Advocate, Delphi, Peer Review, Debate) 📉 Smart model cascading: start with cheap models, escalate on low confidence, per-conversation cost caps 🐳 One Docker image for both MongoDB and PostgreSQL. Switch databases with one env var 🧩 Quarkus SDK: @Inject EddiClient, Dev Services auto-start EDDI in dev mode 🎨 React 19 Manager: full admin UI rewrite, 11 languages, live log streaming, secrets vault UI ✅ 2,400+ tests, the most thoroughly tested release to date
Everything is config-driven. Agent behavior lives in versioned JSON documents. Change routing, responses, or API calls without redeployment, without recompilation. That's not a feature; that's the architecture.
📦 Open source · Apache 2.0 · Red Hat-certified Docker image
→ GitHub: https://github.com/labsai/EDDI → Docs: https://docs.labs.ai/ → Website: https://eddi.labs.ai/