VoiceRun is a serverless enterprise voice platform

Our event-driven paradigm accelerates how enterprises build, deploy, measure, and optimize voice applications with speed, scale, and security.

Code First development

Describe what you want, wire in your systems, and ship versioned deploys with confidence. Build your demo with just a sentence. Then take to your repo, CLI, and IDE to perfect it.

State-of-the-art models

Simply orchestrate between the best model providers, whether for speech-to-text, text-to-speech, or LLM. Be ready to swap from today's SOTA to tomorrow's in an instant without losing native latency optimizations.

Extensible architecture

Pick the design pattern that best suits your use case.

A single reasoning model handles dialogue and actions end-to-end. Lowest complexity and fastest to ship; great for narrow use cases and pilots. As use case complexity growths, context length and model intelligence demand increases until latency, cost, and consistency become blockers.

Flexible deployment

Cloud

Global scale

Instant global availability and edge latency

Autoscaling with no cold starts

Managed security, monitoring, and updates

SOC2/ISO-ready posture out of the box

On-Prem

Max control

Deploy inside your VPC or air‑gapped network

Full data residency and fine-grained control

Offline-capable with the same APIs and SDKs

Custom SLAs, SSO/SAML, and private networking

Simple front-end

Our Telephony

Minutes to live

Managed numbers with global reach

Provision in seconds — no carrier setup

Autoscaling, monitoring, and failover

Secure by default with recording controls

BYO Telephony

Max control

Connect your own Twilio or CAAS

Keep contracts, numbers, and pricing

SIP/PSTN routing and fine‑grained control

Private networking and compliance options

In‑App

Web & mobile

React/Next.js and native iOS/Android SDKs

Realtime mic capture, playback, and stats

Drop‑in components and simple APIs

Works in browsers and mobile apps

Custom measurements

Centralize call logs, user turns, and outcomes. Analyze funnels and identify friction with real transcripts. Instrument your agent directly in code to unlock downstream analytics and iteration.

agent.py

import asyncio
import datetime
import json
from typing import Optional

from primfunctions.events import (
    Event, LogEvent, TextToSpeechEvent, TextEvent, StartEvent
)
from primfunctions.context import Context

# Application constants
SYSTEM_PROMPT = "You are a data-driven assistant that runs quick experiments."
VOICE = "lyric"

# ----- Handlers -----
async def handler(event: Event, context: Context):
    if isinstance(event, StartEvent):
        # 
        # A/B test for welcome message
        context.add_test("welcome_variant", {
            "A": 0.5,  # Control
            "B": 0.5   # Treatment
        }, stop={
            "iterations": 1000,
            "confidence": 95,
            "target_outcome": "conversion_rate",
            "stop_on": 2,
            "default": "A",
            "notify": ["analytics@company.com"]
        })

        selected = context.assign_variant("welcome_variant")
        if selected == "A":
            yield TextToSpeechEvent("Welcome back. Ready to analyze conversions?", voice=VOICE)
        else:
            yield TextToSpeechEvent("Welcome! Want to try a quick experiment to lift conversions?", voice=VOICE)
        # 

    elif isinstance(event, TextEvent):
        user_text = event.data.get("text", "").strip()
        if not user_text:
            return

        # Fetch prior metrics
        current_rate = context.get_state().get("conversion_rate", 0.32)
        last_updated: Optional[str] = context.get_state().get("last_updated")

        # Respond with current numbers
        reply = f"Current conversion rate is {current_rate:.2%}. Last updated: {last_updated or &#39;n/a&#39;}."
        yield TextToSpeechEvent(reply, voice=VOICE)

        # Record we discussed metrics
        yield LogEvent({"event": "metrics_discussed", "rate": current_rate})

        #  

        # Persist the time this was evaluated
        context.set_state({"last_updated": datetime.datetime.utcnow().isoformat()})

Rapid experimentation

Run A/Bs on prompts, voices, and policies. VoiceRun will automatically split and execute your experiment.Capture outcomes per session to compare win rates and deploy the best variant confidently.

agent.py

import asyncio
import datetime
import json
from typing import Optional

from primfunctions.events import (
    Event, LogEvent, TextToSpeechEvent, TextEvent, StartEvent
)
from primfunctions.context import Context

# Application constants
SYSTEM_PROMPT = "You are a data-driven assistant that runs quick experiments."
VOICE = "lyric"

# ----- Handlers -----
async def handler(event: Event, context: Context):
    if isinstance(event, StartEvent):
        # 
        # A/B test for welcome message
        context.add_test("welcome_variant", {
            "A": 0.5,  # Control
            "B": 0.5   # Treatment
        }, stop={
            "iterations": 1000,
            "confidence": 95,
            "target_outcome": "conversion_rate",
            "stop_on": 2,
            "default": "A",
            "notify": ["analytics@company.com"]
        })

        selected = context.assign_variant("welcome_variant")
        if selected == "A":
            yield TextToSpeechEvent("Welcome back. Ready to analyze conversions?", voice=VOICE)
        else:
            yield TextToSpeechEvent("Welcome! Want to try a quick experiment to lift conversions?", voice=VOICE)
        # 

    elif isinstance(event, TextEvent):
        user_text = event.data.get("text", "").strip()
        if not user_text:
            return

        # Fetch prior metrics
        current_rate = context.get_state().get("conversion_rate", 0.32)
        last_updated: Optional[str] = context.get_state().get("last_updated")

        # Respond with current numbers
        reply = f"Current conversion rate is {current_rate:.2%}. Last updated: {last_updated or &#39;n/a&#39;}."
        yield TextToSpeechEvent(reply, voice=VOICE)

        # Record we discussed metrics
        yield LogEvent({"event": "metrics_discussed", "rate": current_rate})

        # context.set_outcome("conversion_rate", current_rate + 0.1)

        # Persist the time this was evaluated
        context.set_state({"last_updated": datetime.datetime.utcnow().isoformat()})

Build your first voice agent today.

Spin up an agent, connect telephony or the React SDK, and ship a pilot in hours.

Get Started