GenXAI is organized around a layered architecture that keeps orchestration, agent execution, memory, and infrastructure concerns separated.

Layers

• Presentation: CLI/SDK/API and (enterprise) no-code Studio
• Orchestration: graph engine, flow control, state manager
• Agent layer: agent runtime, memory system, tool registry
• Communication: message bus, event stream, pub/sub patterns
• Infrastructure: LLM providers, vector DBs, observability

Core components

Graph Engine

The heart of GenXAI: execute workflows as directed graphs.

• Conditional edges and dynamic routing
• Parallel + sequential execution
• Cycles/loops and subgraph composition

Memory System

Layered memory inspired by cognition:

• Short-term + long-term
• Episodic + semantic
• Procedural + working memory

Tool System

An extensible registry for built-in and custom tools, plus integrations via MCP.

Read the full architecture document

The complete architecture lives in the repository at:

• ARCHITECTURE.md