AI Platform · Open source, MIT

GenXAI

A graph-native agentic AI framework.

GenXAI is an open-source, graph-native agentic AI framework for Python, released under the MIT licence. Most agent frameworks model a workflow as a chain — one step, then the next. Real work does not behave that way. It branches on a condition, runs three things at once and waits for all of them, loops until a check passes, and sometimes stops early. GenXAI treats a workflow as a directed graph, which turns all of that from a workaround into ordinary usage. It is free and MIT licensed, with no commercial edition held back.

Read the docs Request a demo

Who it is for

Built for a specific job

AI engineers

Building multi-agent systems that have to run unattended — with retries, rate limits and an audit trail, not a demo script.

Platform teams

Standardising how agents get built across an organisation, with RBAC and shared connectors instead of nine bespoke integrations.

Researchers

Trying coordination patterns — critic review, ensemble voting, map-reduce — without rebuilding the plumbing underneath each one.

What it does

Capabilities

Workflows are graphs

Conditional edges, routing, loops, and parallel branches that join with correct semantics. The topology is the program.

Memory in layers

Short and long-term, episodic, semantic, procedural and working memory as distinct layers, not one undifferentiated context window.

45+ built-in tools

Search, files, code execution, HTTP and data handling out of the box, with a clean interface for adding your own.

MCP in both directions

GenXAI is a Model Context Protocol client and an MCP server, so it consumes external tools and exposes its own to other systems.

14 coordination patterns

Critic review, ensemble voting, map-reduce and others available as named patterns you compose rather than reimplement.

A real production runtime

Connectors, triggers, RBAC, audit logging and rate limiting live in the framework, not in glue code wrapped around it.

How it works

End to end

Define the graph

Nodes are agents, tools or plain Python functions. Edges carry the conditions that decide what runs next.

Attach memory and tools

Each agent gets the memory layers and capabilities it needs. Nothing is global by default.

Pick a provider

OpenAI, Anthropic, Google, Cohere or a local model through Ollama — swappable without rewriting the graph.

Run it in production

Triggers fire the graph, RBAC governs who may do what, audit logging records what actually happened.

Under the hood

Stack and standards

Python
MIT licence
OpenAI · Anthropic · Google · Cohere · Ollama
MCP client and server
45+ tools
14 coordination patterns

Pricing

What it costs

GenXAI is completely free and open source under the MIT licence. There is no commercial edition, no paywalled module and no seat count — the framework you read on GitHub is the whole thing.

Talk to us genxai-framework.com

Questions

Frequently asked questions

What is GenXAI?

GenXAI is an open-source agentic AI framework for Python. It models a workflow as a directed graph rather than a linear chain, so conditional routing, parallel branches with correct join semantics, and loops are first-class features rather than workarounds.

Is GenXAI free to use commercially?

Yes. GenXAI is released under the MIT licence. There is no paid edition, no paywalled module and no seat limit — the framework published on GitHub is the complete product.

Which LLM providers does GenXAI support?

OpenAI, Anthropic, Google, Cohere, and local models through Ollama. The provider is swappable without rewriting the graph.

Does GenXAI support the Model Context Protocol?

Yes, in both directions. GenXAI acts as an MCP client, consuming tools from external MCP servers, and as an MCP server, exposing its own tools to other systems.

How is GenXAI different from chain-based agent frameworks?

Chain-based frameworks run one step after another. Because GenXAI workflows are directed graphs, branching on a condition, running several branches at once and joining them correctly, and looping until a check passes are ordinary usage rather than escape hatches.

What memory does GenXAI provide?

Five distinct layers — short-term, long-term, episodic, semantic and procedural, plus working memory — rather than one undifferentiated context window.

Also from Genexsus

See GenXAI on your own data

A demo with your numbers in it tells you more than any page can. If it turns out GenXAI is not the right fit, we will tell you that too.

Request a demo genxai-framework.com