Docs / A2A

DeepTap as an A2A peer.

Agent-to-Agent gives one agent a uniform way to call another agent's published operations. DeepTap ships an A2A endpoint so any A2A-capable agent can request retrieval, research, or fact lookup without writing DeepTap-specific code.

Agent card

The A2A agent card lives at https://api.deeptap.ai/.well-known/agent.json. It declares the operations DeepTap exposes, their input and output schemas, the supported auth modes, and the streaming capabilities.

{
  "agent": {
    "name": "deeptap",
    "version": "1.0",
    "capabilities": {
      "streaming": true,
      "long_running": true
    },
    "operations": [
      "search",
      "research",
      "cache_lookup"
    ],
    "auth_modes": ["api_key", "x402", "mpp", "truecom"]
  }
}

Supported operations

  • search: depth-1 or depth-2 retrieve.
  • research: depth-3 iterate plus synthesize, streamed.
  • cache_lookup: read-only cache hit query.

Authentication

A2A bearer-token auth maps directly to DeepTap's API key mode. For machine-to-machine commerce, the calling agent can negotiate x402 or MPP via the standard A2A negotiation step. TrueCom-signed receipts are issued by the auth layer regardless of the negotiated mode. See Auth and payment.

Worked example, an agent calling DeepTap

Calling agent invokes the search operation with depth 2:

POST https://api.deeptap.ai/a2a/v1/operations/search
Authorization: Bearer dt_live_...
Content-Type: application/json

{
  "input": {
    "query": "What is the legal status of AI training on copyrighted data in the EU as of mid-2026?",
    "depth": 2
  },
  "stream": false
}

DeepTap responds with the saturated facet ledger and the facts:

200 OK
Content-Type: application/json

{
  "operation_id": "op_01HXY...",
  "status": "complete",
  "output": {
    "depth": 2,
    "rounds_executed": 2,
    "stop_reason": "facet_ledger_saturated",
    "facts": [
      {
        "claim": "...",
        "confidence": 0.88,
        "decay_class": "fast",
        "source_url": "https://...",
        "firewall_verdict": "ok"
      }
    ]
  },
  "receipt": {
    "credits_charged": "{{TBD-depth-2-credits}}",
    "issued_at": "2026-04-24T12:34:56Z"
  }
}

Streaming research

A2A's streaming transport pairs naturally with depth-3 research. The calling agent opens the operation with stream: true and receives progress events until the stopping AI marks the question answered.

POST https://api.deeptap.ai/a2a/v1/operations/research
{
  "input": { "query": "...", "depth": 3 },
  "stream": true
}

// Server-Sent Events
event: round_started
data: { "round": 1 }

event: round_complete
data: { "round": 1, "facets_filled": 3 }

event: stopped
data: { "stop_reason": "stopping_ai_satisfied" }

event: synthesis
data: { "report_html": "..." }

Why A2A matters here

An agent runtime that already speaks A2A can plug DeepTap in alongside other A2A peers (R1 agents, RelayGate ContextWorkers, internal tools) without writing a custom client. The agent card carries everything needed for capability discovery and auth negotiation.