Convergio: Imagine having a super-intelligent command center with 40+ AI specialists. You are the CEO, they are your best team.

[Roberdan]

Convergio.io

Convergio turns a team’s intent into coordinated execution by orchestrating 40+ specialized AI agents. You set the goal; the system assembles experts (finance, architecture, PM, security, design, analytics) to plan, decide, and deliver.

🗞️ What this is (and why it matters)

• Convergio is a unified, production‑leaning reference implementation of a multi‑agent platform built on top of Microsoft AutoGen. It is one of the first large‑scale, end‑to‑end implementations of AutoGen in an open repository, showing how to compose dozens of purpose‑built agents into reliable workflows. See Microsoft AutoGen on GitHub: https://github.com/microsoft/autogen
• Our north star is the Agentic Manifesto and inclusive design. Start here:
  • Agentic Manifesto: AgenticManifesto/AgenticManifesto.md
  • What is Convergio: AgenticManifesto/WhatIsConvergio.md
  • Built for Mario — AI‑First Interface: AgenticManifesto/💜 For Mario - AI-First Interface.md
• Communication meets engineering: this README is both a product primer and a practical guide to run the stack locally.

---

🏗️ Architecture (truth, no buzzwords)

• Backend: FastAPI (Python 3.11), SQLAlchemy 2.x (async), PostgreSQL, Redis, Prometheus metrics
• Frontend: SvelteKit + TypeScript + TailwindCSS (dev server on port 4000, proxy to backend 9000)
• AI Orchestration: Multi‑agent system using Microsoft AutoGen 0.7.x (autogen-core, autogen-agentchat, autogen-ext)
• Streaming & coordination: internal streaming orchestrator with typed protocols and runners
• Security: security headers, CORS, JWT scaffolding (RS256), optional rate‑limit middleware (currently disabled in code)

No Docker/Kubernetes manifests are provided in this repo. Local developer setup runs with system Postgres + Redis or containers you manage yourself.

Key capabilities

• Multi‑agent conversations with tool use, memory, and specialization
• GraphFlow for generating workflows from natural language
• Cost tracking, analytics endpoints, and observability hooks
• Vector search utilities for RAG‑style enrichment