Front Door

The problem

Customer agents already contact businesses through channels built for people. They place phone calls with synthesized voices, fill in web forms, and type into chat widgets. Each step turns structured intent into speech or prose and back again, which adds delay, transcription errors, and per-minute cost on both sides.

Neither side can confirm who the other represents. If your company runs an AI service agent, many of these conversations are already agent-to-agent. They are just happening over the slowest channel available.

How it works

1. Emissar publishes an A2A Agent Card for your business and hosts the endpoint behind it.
2. Incoming agents are identified: by signed requests on the web, and through Resolve when they start from a phone number.
3. Verified agents send structured tasks, such as a refund, an order status check, or a booking, which are routed to your existing service agent or systems.
4. Anything outside your policy goes to Handoff with the full exchange attached.
5. People's own AI assistants reach the same entrance, under the same checks, and are told when a request changes, so the assistant doesn't have to keep asking.

Open standards

A2A defines how two agents exchange tasks once they are connected, and how an agent is told when a task changes. Front Door is the hosted side of that connection for businesses that don't want to run agent infrastructure themselves. Assistants that speak MCP, the Model Context Protocol, use the same entrance through an MCP endpoint, and personal agents that speak PACT, the Personal Agent Consent & Trust Protocol, hold a conversation with it in plain words.

Availability

Accepting design partner applications. If your company already runs an AI service agent and sees automated traffic on its phone or chat channels, apply to the design partner program. A demo with made-up data answers A2A and MCP requests at demo.frontdoor.emissar.ai: a freight company that takes quote requests, order status checks, delivery exceptions and refund requests, the last the way a person's assistant would send one. Personal agents that speak PACT can talk with the same demo at frontdoor.emissar.ai/a2a/demo.

Try the demo

A demo Front Door for a made-up freight company answers A2A requests at demo.frontdoor.emissar.ai. It takes quote requests, order status checks and delivery exceptions, asks for missing details, and holds large quotes for review. Every answer is demo data. Its card is at /.well-known/agent-card.json; you can also point the A2A endpoint tester at it.

curl -s https://demo.frontdoor.emissar.ai/a2a/v1 \
  -H 'Content-Type: application/json' -H 'A2A-Version: 1.0' \
  -d '{"jsonrpc":"2.0","id":1,"method":"SendMessage","params":{"message":{
        "role":"ROLE_USER","messageId":"m1",
        "parts":[{"data":{"orderId":"PO-1001"},"mediaType":"application/json"}],
        "metadata":{"skill":"order_status"}}}}'

# The task comes back TASK_STATE_COMPLETED with an artifact:
# "Demo data. Order PO-1001: Out for delivery, last scan at Chicago, IL,
#  about 4 business days to delivery. Not a real shipment."

Don't send real personal or business data to the demo. What it stores and for how long is in the trust center.

Integration guide

For the engineers who would connect your systems: the Front Door integration guide shows what your systems receive, what they send back, and how each request is checked and reviewed.

Questions

Do we need to replace our current AI service agent?

No. Front Door sits in front of the agent or systems you already run and passes them structured tasks.

Does this change anything for human callers?

No. People reach you through the same phone numbers and chat channels as before.

Can people's own AI assistants use it?

Yes. An assistant acting for one of your customers sends the same requests, with the same checks and your review rules, and Front Door tells it when the request changes. Who the request is for arrives as the assistant gave it, so you verify the person before acting, as you would on a call.