fcan

# Function Calling A2A Network > A framework that empowers Ollama LLM agents in A2A networks with function calling. ## Overview In its current state, this project is a prototype that tries to make the most of function calling in A2A agents. It implements the A2A protocol (minus stuff like streaming responses and input/output negotiation), and sets up LLM agents with system prompts that enable function calling. Currently, the prompts for the LLM agents have been written and tested with Gemma 3 models (4B and 27B) running offline with [`ollama`](https://github.com/ollama/ollama) to talk to the LLMs, it should be easy to add support for other models in the future. The next steps for this project involve adding an orchestration agent, which discovers and delegates tasks to the correct agents based on their skills as advertised in their agent cards. This agent may be an `A2AServer` itself, and the delegation of tasks may be done via function calls. ## Getting Started To get started with this, clone the repository: ```bash git clone https://github.com/gamemaker1/fcan ``` The project uses [`mise`](https://github.com/jdx/mise) for tools and [`uv`](https://github.com/astral-sh/uv) for package management. If you do not have `mise` installed, make sure you have the tools in [`mise.toml`](mise.toml) installed on your system. Make sure you also have [`ollama`](https://github.com/ollama/ollama) up and running on your machine, with the `gemma3` model downloaded. ## Examples > The instructions for running the examples are given in the respective readmes. The [`multi-turn`](examples/multi-turn) example demonstrates how a multi-turn conversation with an agent takes place via the A2A protocol, with the agent making use of the functions provided to it and requesting input from the user when it is not provided. The [`multi-agent`](examples/multi-agent) example demonstrates how to setup and run multiple agents on different ports. It is incomplete, since it needs an orchestration agent to tie it all together.