protoAgent
A lean, A2A-native agent template built on LangGraph, meant to be forked: the boilerplate for the A2A protocol, tracing and the release pipeline stays fixed while a new agent's persona, prompts and tools become the fork's actual changes.
protoAgent ships a small core that runs one agent or a fleet, driven from a console, an OpenAI-compatible API, or the A2A protocol itself. A2A server handling and the LangGraph agent runtime are stable across every fork, so building a new agent is close to rewriting a persona file, a prompts file and a tools file rather than the whole codebase.
A reference fork is provided as a filled-in example: an autonomous portfolio-manager agent with its own persona, A2A skills and project registry. The setup wizard walks through cloning, syncing dependencies and opening a browser tab with no Docker or manual renaming required to get a first agent talking.
An LLM gateway module points an OpenAI-compatible client at LiteLLM so models can be swapped by editing configuration rather than forking code. A delegation mechanism routes sub-tasks to other agents or endpoints over A2A, OpenAI or ACP through a built-in, hot-swappable registry with a health prober, including spawning a CLI coding agent over the Agent Client Protocol.