ovos-a2a-solver-plugin

An OVOS ChatEngine plugin that lets an ovos-persona delegate its reasoning to any external agent that speaks the Agent2Agent (A2A) protocol.

Implements the consumer side of OpenVoiceOS/ovos-persona#164.

What is A2A?

Agent2Agent (A2A) is an open protocol that lets software agents interoperate regardless of their internal implementation. An A2A-compliant server:

1. Exposes a discovery document at GET /.well-known/agent.json (the agent card) that describes its identity, skills, and capabilities.
2. Accepts tasks at its root URL as JSON-RPC 2.0 requests (tasks/send for blocking, tasks/sendSubscribe for SSE streaming).
3. Returns structured responses with typed artifact parts (text, file, data).

This plugin implements the client (consumer) side only — it calls remote A2A agents from within an OVOS persona.

Installation

pip install ovos-a2a-solver-plugin

Requires Python ≥ 3.10 and ovos-plugin-manager >= 2.3.0a1.

Configuration

Add the plugin as the engine for an ovos-persona persona:

# ~/.config/mycroft/personas/my-a2a-persona.yaml
name: my-a2a-persona
engine: ovos-a2a-solver
engine_config:
  agent_url: "https://my-a2a-agent.example.com"   # required
  auth_header: "Bearer <token>"                    # optional
  timeout: 60                                      # seconds (default 60)
  streaming: false                                 # set true to use SSE streaming

Persona wiring example

from ovos_persona import PersonaService

# PersonaService loads persona YAML files automatically.
# Once the plugin is installed, setting engine: ovos-a2a-solver in the
# persona YAML is all that is needed.
svc = PersonaService(config={})
reply = svc.chat("What is the capital of Portugal?", persona="my-a2a-persona")
print(reply)  # "Lisbon."

Or use the engine directly:

from ovos_a2a_solver import A2AChatEngine
from ovos_plugin_manager.templates.agents import AgentMessage, MessageRole

engine = A2AChatEngine(config={
    "agent_url": "https://my-a2a-agent.example.com",
})

messages = [
    AgentMessage(role=MessageRole.SYSTEM, content="You are a helpful assistant."),
    AgentMessage(role=MessageRole.USER, content="Hello!"),
]
reply = engine.continue_chat(messages)
print(reply.content)

For streaming:

engine = A2AChatEngine(config={
    "agent_url": "https://my-a2a-agent.example.com",
    "streaming": True,
})

for chunk in engine.stream_tokens(messages):
    print(chunk, end="", flush=True)

A2AClient — low-level usage

The A2AClient class can be used independently of OPM:

from ovos_a2a_solver import A2AClient

with A2AClient("https://my-a2a-agent.example.com") as client:
    card = client.fetch_agent_card()
    print(card.name, card.skills)

    answer = client.send_task(
        "Summarise the A2A spec in one sentence.",
        session_id="my-session",
    )
    print(answer)

Protocol notes

Agent-card discovery

On __init__ the engine fetches GET {agent_url}/.well-known/agent.json and logs the agent's name. A failure is non-fatal — the engine continues and will attempt tasks anyway (useful when the card endpoint is behind auth).

Message history

OVOS ChatEngine receives the full conversation list. The plugin maps it to A2A history as follows:

Streaming

When streaming: true the plugin calls tasks/sendSubscribe and yields text chunks via SSE. If the server's agent card reports capabilities.streaming: false the plugin falls back to blocking tasks/send and warns.

Development

git clone https://github.com/TigreGotico/ovos-a2a-solver-plugin
cd ovos-a2a-solver-plugin
pip install -e ".[test]"
pytest test/ -v

Credits

Developed by TigreGótico for OpenVoiceOS.

This project was funded through the NGI0 Commons Fund, a fund established by NLnet with financial support from the European Commission's Next Generation Internet programme, under the aegis of DG Communications Networks, Content and Technology under grant agreement No 101135429.

License

Apache License 2.0 — see LICENSE.