A2A Protocol¶

Lango supports the A2A protocol for inter-agent communication. This allows a Lango instance to advertise its capabilities via an Agent Card and to discover and integrate external A2A-compatible agents as sub-agents.

Overview¶

The A2A implementation has two sides:

1. Server -- Exposes this Lango agent as an A2A service with a discoverable Agent Card
2. Client -- Discovers and integrates remote A2A agents into the local orchestrator tree

graph LR
    subgraph Local Lango
        O[Orchestrator] --> L[Local Sub-Agents]
        O --> RA[Remote Agent Proxy]
    end

    subgraph Remote Service
        AC[Agent Card<br/>/.well-known/agent.json]
        RS[Remote Agent]
    end

    RA -- "A2A Protocol" --> RS
    RA -. "discovery" .-> AC

    style AC fill:#e67e22,color:#fff
    style RA fill:#e67e22,color:#fff

Agent Card¶

When A2A is enabled, Lango serves an Agent Card at /.well-known/agent.json. The card describes the agent's name, description, URL, and skills.

Card Structure¶

{
  "name": "lango-assistant",
  "description": "Lango AI Assistant",
  "url": "https://your-host:18789",
  "skills": [
    {
      "id": "lango-orchestrator",
      "name": "lango-orchestrator",
      "description": "Lango Assistant Orchestrator",
      "tags": ["orchestration"]
    },
    {
      "id": "operator",
      "name": "operator",
      "description": "System operations: shell commands, file I/O, and skill execution",
      "tags": ["sub_agent:lango-orchestrator"]
    }
  ]
}

Skills are automatically derived from the agent tree. The root agent is listed with the orchestration tag, and each sub-agent is listed with a sub_agent:<parent> tag.

Remote Agents¶

Remote A2A agents are discovered by fetching their Agent Card from a configured URL. Each remote agent is integrated as a sub-agent in the multi-agent orchestrator.

Discovery Process¶

1. Lango fetches the Agent Card from the configured agentCardUrl
2. The ADK remoteagent.NewA2A() creates a proxy agent from the card
3. The proxy is added to the orchestrator's sub-agent list
4. The orchestrator can delegate tasks to the remote agent via transfer_to_agent

Graceful Degradation¶

If a configured remote agent cannot be constructed during startup, Lango logs a warning and skips that remote. Successfully loaded remotes are still attached to the orchestrator, and the rest of the local agent tree continues to function normally. Remote loading failures do not block startup, but missing configuration such as an empty agentCardUrl is surfaced in the startup warning instead of being silently ignored.

WARN  load remote agent  name=weather-agent  url=https://weather.example.com/.well-known/agent.json  error=connection refused
WARN  load remote A2A agents  error="remote agent \"review-agent\" missing agentCardUrl"

Configuration¶

Settings: lango settings → A2A Protocol

{
  "a2a": {
    "enabled": true,
    "baseUrl": "https://your-host:18789",
    "agentName": "lango-assistant",
    "agentDescription": "Lango AI Assistant",
    "remoteAgents": [
      {
        "name": "weather-agent",
        "agentCardUrl": "https://weather.example.com/.well-known/agent.json"
      },
      {
        "name": "code-review-agent",
        "agentCardUrl": "https://review.example.com/.well-known/agent.json"
      }
    ]
  }
}

Each remote agent entry requires:

lango agent list shows configured remote entries. Use lango agent list --check or lango doctor to validate reachability before relying on a remote as a delegation target.

Setup¶

Enable via Settings¶

Enable the A2A protocol through the full settings editor or config import/export workflow:

lango settings

Add Remote Agents¶

Remote agents are configured via the config file. Use the config export/edit/import workflow:

# Export current config
lango config export default > config.json

# Edit to add remote agents under a2a.remoteAgents
# ...

# Import updated config
lango config import config.json --profile default

A2A-over-HTTP vs A2A-over-P2P¶

Lango supports two modes of inter-agent communication:

Both modes can be enabled simultaneously. A2A-over-HTTP is simpler to set up for public agents, while A2A-over-P2P provides stronger security guarantees and works without centralized infrastructure.

See P2P Network for P2P-specific documentation.