100% open source · MIT licensed
Build, run, and evolve multi-agent workflows with graph orchestration, multi-layer memory, and an extensible tool system. Fully open source — production features included.
Why GenXAI
GenXAI combines graph execution, memory, and tools into a framework you can grow from prototype to production.
Model workflows as graphs (not only linear chains). Route dynamically, run in parallel, loop, and compose subgraphs.
Multi-layer memory inspired by human cognition so agents can retain context, learn, and retrieve knowledge efficiently.
Built-in tools, custom tools, and Model Context Protocol integrations so agents can act in real systems.
Everything ships MIT-licensed: connectors, triggers, observability, security, and the no-code Workflow Studio.
How it works
Use graph primitives for routing and coordination patterns. Start simple, then add memory, tools, and production controls when needed.
Declare roles, goals, models, tools, and memory. Keep it explicit and versionable.
Nodes represent agents/tools/conditions. Edges define control flow, routing, and parallelism.
Execute with state, logs, and metrics. Built-in observability and governance keep production runs reliable.
Quick start
import os
from genxai import Agent, AgentConfig, AgentRegistry, Graph
os.environ["OPENAI_API_KEY"] = "sk-..."
classifier = Agent(
id="classifier",
config=AgentConfig(role="Classifier", goal="Categorize customer requests", llm_model="gpt-4"),
)
support = Agent(
id="support",
config=AgentConfig(role="Support Agent", goal="Resolve customer issues", llm_model="claude-3-opus", enable_memory=True),
)
AgentRegistry.register(classifier)
AgentRegistry.register(support)
graph = Graph()
# ... add nodes + edges ...
result = await graph.run(input_data="My app crashed")
FAQ
Yes. The entire framework — including production features like connectors, triggers, observability, and security — is MIT-licensed. There is no commercial edition and nothing behind a paywall.
GenXAI is a Python framework: pip install genxai-framework. Optional extras enable integrations, e.g. pip install "genxai-framework[llm,tools,api]" — available extras are llm, tools, api, storage, observability, connectors, and all.
OpenAI, Anthropic, Google, Cohere, and Ollama. Multi-provider routing lets different agents in the same workflow use different models.
Yes, as both client and server: agents can call tools from any registered MCP server, and GenXAI can expose its own tools over MCP.
GenXAI is Python-first, but it ships with Workflow Studio — a no-code drag-and-drop visual builder where you compose agent/tool pipelines on a canvas and run them with live per-node status.
Workflows are directed graphs: conditional edges and dynamic routing, parallel execution with real join semantics, loops, and nested subgraphs — plus 14 reusable multi-agent coordination patterns like critic review, ensemble voting, and map-reduce.
Follow the getting-started guide, run an example, and evolve into production-grade orchestration patterns.