A2A & Agent Networks

GaaS as the governance control plane for multi-agent systems — covering the A2A Protocol v1.0, Agent Trust Registry, and AP2 Agentic Payment Governance.

When AI agents talk to each other, delegate tasks, or execute financial transactions autonomously, governance becomes non-negotiable. GaaS is an A2A v1.0 agent: every message another agent sends it passes through the same 5-stage governance pipeline as a human-facing action. It is built on the official A2A SDK and tested on every change with the official A2A conformance suite (TCK) at the MUST level, over JSON-RPC — the binding its Agent Card declares.

Agent Card

GaaS publishes its Agent Card at the spec's well-known path, without authentication: GET /.well-known/agent-card.json. (/.well-known/agent.json, the pre-v1.0 path, redirects there.) version is GaaS's own version; the A2A protocol version is on each entry of supportedInterfaces.

{
  "name": "GaaS Governance Agent",
  "version": "0.2.17",
  "supportedInterfaces": [
    {
      "url": "https://api.gaas.is/a2a/v1",
      "protocolBinding": "JSONRPC",
      "protocolVersion": "1.0"
    }
  ],
  "capabilities": {
    "streaming": true,
    "pushNotifications": false,
    "extendedAgentCard": false
  },
  "securitySchemes": {
    "apiKey": {
      "apiKeySecurityScheme": {
        "description": "Your GaaS API key (the same key as the rest of the API).",
        "location": "header",
        "name": "X-API-Key"
      }
    }
  },
  "securityRequirements": [
    {
      "schemes": {
        "apiKey": {}
      }
    }
  ],
  "skills": [
    {
      "id": "governance-check",
      "name": "Governance Check",
      "tags": [
        "governance",
        "compliance",
        "risk",
        "audit",
        "policy"
      ]
    },
    {
      "id": "risk-assessment",
      "name": "Risk Assessment",
      "tags": [
        "risk",
        "scoring",
        "assessment",
        "dimensions"
      ]
    },
    …
  ]
}

Abridged: the live card also carries a description, provider, documentation link, input and output modes, and all 8 skills.

A2A Gateway

The A2A endpoint is JSON-RPC 2.0 at POST /a2a/v1. Every request must carry A2A-Version: 1.0; a request without it (which the spec reads as 0.3) or with another version gets error -32009 (VersionNotSupportedError). Authenticate with your GaaS API key in X-API-Key.

Methods

Tasks belong to the organization whose key created them. Another organization's task id gets -32001 (TaskNotFoundError), the same as an id that does not exist.

Example: govern an action

POST /a2a/v1
X-API-Key: gsk_...
A2A-Version: 1.0
Content-Type: application/json

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "SendMessage",
  "params": {
    "message": {
      "messageId": "msg-0001",
      "role": "ROLE_USER",
      "parts": [{"text": "Send the quarterly revenue report to the external auditor by email"}]
    }
  }
}

The result is a task. Its governance-decision artifact holds GaaS's decision as a data part (verdict, risk_score, reasoning, audit_id, blocking_policies, …), the status message carries the reasoning, and the task metadata carries gaas_audit_id, gaas_intent_id and gaas_verdict. To pick a skill, put "skill": "<id>" in the message metadata or in a data part. Set "gaas.pipelineMode": "shadow" in the message metadata to evaluate without enforcing.

Decisions as task states

An escalated action becomes a review item in your dashboard, and GaaS emails your organization. The task's metadata carries gaas_escalation_id. Send another message on the same task (with its taskId and contextId) to get the outcome: approved or modified → COMPLETED, denied → REJECTED, timed out → per your timeout rule, still waiting → INPUT_REQUIRED again.

Agent Trust Registry

The Agent Trust Registry is GaaS's decentralized reputation system for autonomous agents. Every agent that interacts with the platform accumulates a trust score based on governance outcomes, policy violations, and human review results.

Trust Score

Trust scores range from 0.0 (untrusted) to 1.0 (fully trusted). New agents start at 0.5 and build reputation through approved actions and positive human review outcomes. Violations, blocks, and escalations reduce the score.

A2A Authentication

POST /a2a/v1 accepts:

• GaaS API Key: the standard X-API-Key header — the same key as the rest of the API. This is what the hosted service accepts.
• Bearer JWT: an Authorization: Bearer token signed with the deployment's A2A secret (GAAS_A2A_JWT_SECRET), carrying org_id and optionally agent_id. Only when that secret is configured; the hosted service does not configure it today.

The Agent Card lists exactly the schemes the deployment accepts. OAuth 2.0 client credentials and mutual TLS are not supported.

Governance Proxy

Send an A2A message to POST /a2a/proxy/messages with the target_agent_url it is meant for. GaaS runs it through the five-stage governance pipeline and returns the result as an A2A task. Forwarding approved messages to the target agent is built, not yet switched on: today your orchestrator forwards an approved message itself, and does not forward a blocked or escalated one. POST /a2a/proxy/messages/shadow runs the same evaluation in shadow mode.

POST /a2a/proxy/messages
X-API-Key: gsk_...
Content-Type: application/json

{
  "role": "user",
  "parts": [
    {"type": "text", "text": "Ship order 1042 to the customer's address on file"}
  ],
  "target_agent_url": "https://fulfillment.example.com/a2a"
}

The target agent's URL must be HTTPS on a public host name: GaaS refuses private, loopback and cloud-metadata addresses, and does not follow redirects, whenever it contacts another agent.

Cross-Org Policy Federation

Built, not yet enabled. Federation lets GaaS negotiate governance requirements and check mutual compliance before agents in two organizations work together. The hosted service does not run it yet, so a cross-organization request is governed only by the receiving organization's own policies.

Push Notifications

Not available yet. The Agent Card says "pushNotifications": false, and the push configuration methods return -32003 (PushNotificationNotSupportedError). Poll with GetTask, or stream with SendStreamingMessage / SubscribeToTask.

AP2 Payment Governance

The Agentic Payment Protocol (AP2) extends GaaS with seven payment-specific policies that govern autonomous agent transactions. AP2 is designed around the emerging reality that AI agents will execute payments autonomously on behalf of users — and those payments require the same compliance safeguards as human-initiated transactions.

Tier 1 AP2 Policies (Fast-Fail)

These four policies run first and immediately block invalid or out-of-scope payment requests:

Tier 2 AP2 Policies (Regulatory)

These three policies enforce payment regulation compliance:

Submitting an AP2-governed Payment Intent

POST /v1/intents
X-API-Key: your_api_key

{
  "intent": {
    "agent": {
      "id": "procurement_agent_v1",
      "framework": "custom"
    },
    "action": {
      "type": "TRANSACT",
      "verb": "initiate_payment",
      "target": {
        "type": "ACCOUNT",
        "identifier": "vendor_account_V9912",
        "sensitivity": "REGULATED"
      }
    },
    "payload": {
      "summary": "Pay vendor invoice INV-2026-0044 for SaaS services",
      "content": {
        "amount_usd": 2500.00,
        "currency": "USD",
        "merchant_category": "7372",
        "mandate_id": "mnd_vendor_9912"
      }
    },
    "estimated_impact": {
      "reversible": false,
      "financial_exposure_usd": 2500.00,
      "regulatory_domains": ["PCI-DSS", "PSD2", "AML"]
    }
  }
}

AP2 Mandate Management

Mandates define the authorized scope for agent payments. A mandate specifies which agent may transact, on whose behalf, with which merchants, up to what limits, and for what period.

Mandate Schema

{
  "mandate_id": "mnd_vendor_9912",
  "agent_id": "procurement_agent_v1",
  "authorized_by": "user_cfo_01",
  "scope": {
    "merchant_categories": ["7372", "7371"],
    "max_single_transaction_usd": 5000.00,
    "daily_limit_usd": 10000.00,
    "monthly_limit_usd": 50000.00,
    "hnp_threshold_usd": 10000.00,
    "currency": "USD",
    "geographic_scope": ["US", "EU"]
  },
  "valid_from": "2026-01-01T00:00:00Z",
  "valid_until": "2026-12-31T23:59:59Z",
  "status": "active",
  "cumulative_spend_usd": 2500.00,
  "remaining_daily_budget_usd": 7500.00
}

Related Pages

• Connectors — A2A Registry and AP2 Mandate connectors
• Intent Declaration API — The core governance entry point
• Policy Library — Full policy catalog including AP2 and Tier 4 AI safety policies
• Webhooks — Event-driven notifications for governance decisions
• Authentication — API keys and A2A auth schemes