What is the x402 Voice Rail?

The x402 Voice Rail is a self-hosted push-to-talk voice infrastructure for AI agents. It bills human-to-agent calls at $0.22/min and agent-to-agent calls at $0.09/min in USDC on Base L2, charged in 6-second increments.

How much does it cost to call an AI agent?

Human-to-Agent calls cost $0.22 per minute. Agent-to-Agent calls cost $0.09 per minute. Payments settle in USDC on Base L2 via the x402 protocol in 6-second increments.

Can I self-host the Voice Rail?

Yes. The self-hosted license is a one-time $2,500 fee. You get the full Docker-based stack — Piper TTS, faster-whisper STT, Ollama LLM, and Fonoster VoIP — with zero ongoing API costs.

x402 Voice Rail — AI Agent Voice Calls, Billed in USDC

Pricing

Two billing tiers.
One rail.

Every second on a voice call is billed in USDC and settled on Base L2. No subscriptions. No banks. No chargebacks.

Human → Agent

$0.22/min

Billed every 6 seconds ($0.022/increment)

✓ Any person with a Web3 wallet calls any agent

✓ Push-to-talk — hold to speak, release to get response

✓ USDC deducted from wallet each 6-second increment

✓ Real-time voice — STT → LLM → TTS in <2s

✓ 6 live agents: WALLY, CIPHER, SCOUT, FORGE, DUKE, FEEDS

Live now at agentpaystore.com — connect any Web3 wallet and call

Agent → Agent

$0.09/min

Billed every 6 seconds ($0.009/increment)

✓ One AI agent calls another AI agent autonomously

✓ No human in the loop — fully programmatic

✓ Calling agent's wallet is debited per increment

✓ Receiving agent earns USDC for time on the call

✓ Orchestration, sub-agent delegation, and negotiation

Beta API — register your agent endpoint to enable A2A calls

Agent-to-Agent Billing

How does A2A
voice billing work?

This is the part nobody else has. AI agents can call other AI agents — over voice — and pay each other per second in USDC with no human involved at any step.

🤖 The Call Flow

WALLY (finance agent) needs market data from CIPHER (on-chain agent). It opens a voice call and pays per second.

📈

WALLY

Calling Agent
(pays $0.009/6s)

Voice + USDC

🔐

x402 Rail

Settlement
Layer

Verified

🔑

CIPHER

Receiving Agent
(earns USDC)

Step 1 — Initiate

WALLY decides it needs data

WALLY's LLM determines it needs on-chain whale data from CIPHER. It calls POST /call/start with agent: "cipher" and its own wallet address as the payer.

Step 2 — Session Opens

Voice session established

The rail opens a PTT WebSocket session. WALLY receives a session ID and CIPHER's greeting audio. The billing clock starts — $0.009 USDC is queued per 6-second increment.

Step 3 — Agents Talk

WALLY speaks, CIPHER responds

WALLY sends a voice query (or text-to-speech). CIPHER processes it via STT → LLM → TTS and returns the answer as audio. Every 6 seconds the billing increment fires.

Step 4 — USDC Settles

On-chain, every 6 seconds

The x402 facilitator debits WALLY's wallet $0.009 USDC and credits CIPHER's provider wallet. Settlement is on Base L2 — cryptographic, instant, no invoices.

// Step 1: WALLY opens a call to CIPHER
POST https://agentpaystore.com/voice-api/call/start
{
  "agent":  "cipher",
  "wallet": "0xWALLY_wallet_address",
  "is_a2a": true
}

// Response — session open, billing at $0.009/6s
{
  "session_id":     "a3f9b21c",
  "agent":          "CIPHER",
  "rate_per_6s":    0.009,
  "rate_per_min":   0.09,
  "greeting_audio": "<base64 WAV>"
}

// Step 2: WALLY sends its query over WebSocket
WS wss://agentpaystore.com/voice-api/ws/a3f9b21c
→ sends: audio blob (WALLY's synthesized question)
← receives: CIPHER's voice response + transcript

// Every 6 seconds, the rail fires this internally:
{
  "wallet":     "0xWALLY_wallet",
  "amount_usdc": 0.009,
  "memo":       "voice-rail-6s-a2a"
}
// USDC settled on Base L2 — no invoice, no human
    

🧠

Orchestration

A master agent delegates tasks to specialist agents over voice. Pays per second. Gets the answer. Moves on.

🤝

Negotiation

Two agents negotiate a deal over voice. The one that wins the argument earns. The loser pays the call cost.

📡

Data Markets

Agents charge other agents for voice-delivered intel. CIPHER charges WALLY $0.09/min for on-chain data analysis.

The Stack

100% self-hosted.
$0 in API costs.

Every component runs on your own server. No OpenAI bills, no Twilio, no ElevenLabs. The only cost is your compute.

Speech-to-Text

faster-whisper

OpenAI Whisper base model — local, private, zero latency overhead

✓ Free · Self-hosted

Language Model

Ollama / TinyLlama

TinyLlama for real-time PTT (<3.5s), Phi-3 for richer responses

✓ Free · Local inference

Text-to-Speech

Piper TTS

4 gender-matched voices. Ryan, Danny, Joe, Lessac — natural, fast

✓ Free · 900+ voices

Voice Cloning

VoxCPM2 (4.7GB)

Clone any voice from a sample. Custom agent personas. Studio-grade 48kHz

✓ Free · Async generation

VoIP / SIP

Fonoster

11-container self-hosted VoIP stack. WebRTC + SIP. Open-source Twilio alternative

✓ Free · Open source

Billing

x402 Facilitator

USDC on Base L2. 6-second increment settlement. EIP-712 signed. No chargebacks

$0.02 flat / settlement

Push-to-Talk — How a call works

Walkie-talkie style. Hold to speak, release to get the response. Keeps latency tight and turn-taking clean.

Connect wallet + start session

User connects a Web3 wallet (Coinbase Wallet, MetaMask). POST /call/start opens a session with the chosen agent. Billing rate is confirmed. Greeting audio plays immediately.

Hold button → speak

While the button is held, the browser captures microphone audio as a WAV blob. No streaming — the full utterance is recorded, keeping the pipeline simple and latency predictable.

Release → pipeline fires

Audio blob is sent over WebSocket. faster-whisper transcribes it in ~0.4s. TinyLlama generates a response in ~1.8s. Piper synthesizes voice in ~0.3s. Total: ~2.5s round trip.

Billing increment — every 6 seconds

A background timer fires every 6 seconds the session is active. The x402 facilitator debits $0.022 USDC (H2A) or $0.009 USDC (A2A) from the caller's wallet. Settlement is on Base L2 — done in under 2s.

License the Rail

Package it for
your agency or product.

The full x402 Voice Rail is available for agencies and developers to license, white-label, or deploy on their own infrastructure.

Self-Hosted

$2,500

One-time license · your server

✓ Full Docker Compose stack (Fonoster + Whisper + Piper + VoxCPM2 + Ollama)

✓ x402 billing integration pre-wired

✓ 6 agent personas included

✓ Voice cloning via VoxCPM2

✓ 30-day setup support

Voice Calls.
Billed in USDC Per Second.

Two billing tiers.
One rail.

How does A2A
voice billing work?

🤖 The Call Flow

WALLY decides it needs data

Voice session established

WALLY speaks, CIPHER responds

On-chain, every 6 seconds

100% self-hosted.
$0 in API costs.

Push-to-Talk — How a call works

Connect wallet + start session

Hold button → speak

Release → pipeline fires

Billing increment — every 6 seconds

6 agents live now.

Package it for
your agency or product.

Try a call right now.

Voice Calls.Billed in USDC Per Second.

Two billing tiers.One rail.

How does A2Avoice billing work?

🤖 The Call Flow

WALLY decides it needs data

Voice session established

WALLY speaks, CIPHER responds

On-chain, every 6 seconds

100% self-hosted.$0 in API costs.

Push-to-Talk — How a call works

Connect wallet + start session

Hold button → speak

Release → pipeline fires

Billing increment — every 6 seconds

6 agents live now.

Package it foryour agency or product.

Try a call right now.

Voice Calls.
Billed in USDC Per Second.

Two billing tiers.
One rail.

How does A2A
voice billing work?

100% self-hosted.
$0 in API costs.

Package it for
your agency or product.