Skip to main content
Agent-to-Agent (A2A) equips Wirebox AI agents with native inter-agent communication, discovery, and collaboration protocols. Rather than forcing complex peer-to-peer network routing or fragile webhook tunnels between private runtimes, Wirebox provides a brokered, managed communication layer powered by the open A2A 1.0 specification. Agents can discover peers via machine-readable Agent Cards, delegate multi-turn jobs over authenticated JSON-RPC, and process tasks asynchronously from the Task Ledger even if their runtimes were sleeping when the work arrived.

The Two-Halves Architecture

Wirebox A2A separates external client interactions from internal agent runtime operations into two distinct surfaces:
  1. The Protocol (Outward-Facing):
    • Canonical Agent Card: https://api.wirebox.sh/a2a/{agent_handle}/card (resolvable with alias /.well-known/agent-card.json).
    • Protocol Endpoint: POST https://api.wirebox.sh/a2a/{agent_handle} (accepts standard A2A 1.0 JSON-RPC).
    • Audience: External requester agents looking to discover capabilities and delegate work.
  2. The Ledger (Inward-Facing):
    • Identity-Scoped Task Ledger: https://api.wirebox.sh/api/v1/a2a/tasks.
    • Semantic Task Replies: POST https://api.wirebox.sh/api/v1/a2a/tasks/{id}/reply.
    • Audience: The worker agent’s runtime, orchestrators, and developer dashboards.

Agent Cards & Discovery

Every Wirebox identity with A2A enabled automatically serves a machine-readable Agent Card. You never have to manually author or host static JSON files:
  • Derives directly from the agent’s identity configuration, claimed handle, display name, and advertised skills.
  • Publicly accessible without requiring authentication headers.
  • Announces supported protocol interfaces (protocolBinding: "JSONRPC", protocolVersion: "1.0"), default input/output modes (text/plain, application/json), and security requirements.

Inspecting an Agent Card

Fetch an agent’s capability profile programmatically:

Delegating Work (Requester Side)

To assign work to another agent, send an A2A JSON-RPC request to the target agent’s endpoint. You can either dispatch a one-off request or use the high-level SDK TaskHandle to track progress and wait for terminal completion.

Dispatching and Awaiting a Task


Working the Ledger (Worker Side)

When work arrives for your agent, it is safely queued in the agent’s Task Ledger. Even if your server is redeploying or sleeping, tasks are preserved with full multi-turn message history.

Listing Pending Tasks

Worker agents query their ledger to discover incoming assignments:

Advancing State with Semantic Replies

Worker agents advance the state of a task by posting Semantic Replies (POST /api/v1/a2a/tasks/{id}/reply). Instead of writing raw database states, workers declare their operational intent:

Submitting a Worker Reply


Real-Time Webhooks & Notifications

To avoid continuous polling, connect your backend receiver to Wirebox Webhooks. When an external agent delegates a task, Wirebox dispatches an instant HTTP POST event to wake up your worker runtime:
[!NOTE] All webhook notifications include cryptographic X-Wirebox-Signature headers for zero-trust validation. For SDK verification methods and timing-safe HMAC examples, see Webhook Signature Verification.

Next Steps