Agentic AI Architecture — Production AutoGen, MCP, MAF
Ohad Phoenix Oren
Multi-Agent Swarm Orchestration
Built MAF 1.0 with native A2A protocol (Google standard) for multi-agent communication. Designed semantic filtering and density governor for context window optimization. Implemented memory persistence with auto-compression across agent sessions.
Architecture
Multi-provider AI proxy architecture: zero client-side API keys, single backend routing to Claude, GPT, Gemini, Copilot. Agentic loops with native tool use and agent-level access control.
Stack
React 19 · TypeScript · JavaScript · Vite · Python (FastAPI) · Node.js · PostgreSQL · Turso · Railway · A2A protocol (Google standard) · MCP (Model Context Protocol) · Claude API / Anthropic Console · GitHub Copilot multi-model development environment (Claude, GPT, Gemini, Grok)
Like this project
Posted Apr 15, 2026
Multi-agent swarm orchestration with A2A protocol (Google standard), MCP, MAF 1.0. Turso edge DB. PostgreSQL. Zero client-side keys. LLM-agnostic routing across Claude, GPT, Gemini, Copilot.
Likes
0
Views
1
