Building multi-agentic systems with AGNTCY's Application SDK and reference application

Today, we are pleased to announce the launch of the multi-agent software reference application we call coffeeAGNTCY, as well as an Application SDK. 

CoffeeAGNTCY is the fastest way to experiment within the agentic open source ecosystem. See how to build multi-agent software with Model Context Protocol (MCP), Agent2Agent (A2A), and AGNTCY components. It's where “hello world” meets “hello agents.”

The reference application muse is a fictional global coffee enterprise. Open source reference applications are not new. Just like the cloud native Sock Shop began as a simple demo app, then grew into a fully containerized, cloud native reference application, we are starting small with coffeeAGNTCY.

The first version has a simple two-agent system we call Corto, and a full multi-agent simulation with different patterns of communication, which we call Lungo. 

"The coffeeAGNTCY reference application is a great launchpad for us to implement AGNTCY components in our latest project. The easy-to-follow ReadMe and ready-made Docker compose files are super convenient," said Amogh Tarcar, Senior Data Scientist at Persistent.

Watch the overview video below for more details:

The coffeeAGNTCY is built to be extensible for everyone building multi-agent applications and is designed to support ALL open source protocols for a multi-agent system. 

Get started with the Corto demo

We recommend starting with the Corto ReadMe. Here, we demonstrate the integration of an A2A client within a LangGraph workflow with an A2A server agent. It models a simplified agent system that acts as a coffee sommelier.

• The exchange agent acts as a client interface, receiving prompts from the user interface about coffee flavor profiles and forwarding them to the farm agent.
• The farm agent serves as a backend flavor profile generator, processing incoming requests and returning descriptive output.

The user interface forwards all prompts to the exchange’s API, which are then given to a LangGraph which contains an A2A client node. This A2A client node connects to the farm’s A2A server. The underlying A2A transport layer is fully configurable. By default, the system uses AGNTCY's SLIM (Secure Low-latency Interactive Messaging), but it’s built to be interoperable, so you can use NATS, Kafka, or your preferred messaging protocol.

You can use Corto in two ways: Local Python which runs each component directly on your machine or via Docker Compose which will quickly spin up all components as containers. 

Building with the AGNTCY App SDK 

The AGNTCY Application SDK provides a convenient way to integrate with AGNTCY components, beginning with SLIM and later expanding to incorporate many others including Observe-SDK, and Identity. All this while enabling interoperability with agentic protocols such as A2A and Model Context Protocol.

Stay tuned for more coming soon!

Shape the future of open agentic development with AGNTCY

Helping the world build multi-agentic systems with open source solutions is one of our passions here at Outshift. The future doesn’t build itself, and we're excited to create foundational tools to empower technologists to create. And we don’t do it alone: dozens of companies have joined AGNTCY with the same vision, building an open, interoperable internet of agents. To get involved in building AGNTCY with us, contribute to working group meetings and join us on Slack.