hivemind-a2a-agent-plugin

An agent-protocol plugin for HiveMind that connects the hive to external A2A (Agent-to-Agent) agents.

The plugin forwards natural-language queries from hive satellites to a configured A2A server over JSON-RPC 2.0 (tasks/send or tasks/sendSubscribe). It streams the response back to the satellite that sent the query.

What is A2A?

A2A is an open protocol for agent interoperability. An A2A server does two things:

1. It publishes an agent card at GET /.well-known/agent.json. The card describes the server's capabilities, skills, and the URL that accepts tasks.
2. It accepts task requests as JSON-RPC 2.0 at its root URL. A request is either a blocking tasks/send call or a streaming tasks/sendSubscribe call (SSE).

Any compliant A2A server works as a backend for this plugin, including LangChain agents, Google ADK, CrewAI, and custom FastAPI services.

Installation

pip install hivemind-a2a-agent-plugin

The plugin registers itself under the hivemind.agent.protocol entry-point group. HiveMind-core discovers it automatically once it is installed.

Configuration

Add the following to your OVOS / HiveMind config, typically at ~/.config/hivemind/hivemind.conf:

{
  "hivemind": {
    "agent_protocol": "hivemind-a2a-agent-plugin",
    "a2a_agent": {
      "agent_url": "http://localhost:9999",
      "auth_header": "Bearer secret",
      "timeout": 60,
      "streaming": false
    }
  }
}

Hive wiring example

HiveMind master
  └── hivemind-a2a-agent-plugin  ←  loaded as agent_protocol
        └── A2A server  (http://localhost:9999)
              └── your LLM / agent / tool backend

Satellite (voice client, phone, …)
  → "what's the capital of France?"
  → HiveMind master receives utterance
  → plugin forwards to A2A server via tasks/send
  → A2A server responds: "Paris is the capital of France."
  → HiveMind master streams answer back to satellite

Run a minimal FastAPI A2A server and the HiveMind master:

# 1. run the example mock A2A server (also used by e2e tests), from tests/
cd tests && uvicorn e2e.mock_a2a_server:app --port 9999

# 2. configure hivemind-core to use this plugin (see above)
# 3. start hivemind-core
hivemind-core listen

Session and context mapping

The plugin maps the HiveMind session_id directly to the A2A sessionId parameter, so multi-turn conversations keep their context on the A2A server side. The plugin stores no extra state. The A2A server owns the conversation history.

Error handling

The plugin never silences errors. The A2A server may be unreachable, return an empty response, or return a JSON-RPC error object. In each case, the plugin yields a human-readable error string to the satellite so the user always gets a reply.

Development

git clone https://github.com/JarbasHiveMind/hivemind-a2a-agent-plugin
cd hivemind-a2a-agent-plugin
pip install -e ".[dev]"
pytest tests/          # unit tests (no server needed)
pytest tests/e2e/      # e2e tests (spins up a FastAPI mock server in-process)

Related projects

• JarbasHiveMind/HiveMind-core: the HiveMind master this plugin connects to.

Credits

Funded by NGI0 Commons Fund / NLnet under grant agreement No 101135429, through the European Commission's Next Generation Internet programme.