Open core · self-host or run on Unpodv1.0 ↗

Realtime voice + messaging infrastructure

Telephony & Speech Infra for AI Agents.

Telephony and speech delivered as infra - with an open-source CPaaS and a conversation framework on top. Build the agent, self-host the platform, or run the whole stack on Unpod.

Start Building →Read the docs

Developer PlatformTelephony + Speech UnpodOpen-source CPaaS SuperDialogConversation framework

One platform · three layers · text in, text out

What teams ship on Unpod

AI ReceptionistCoding-agent phoneOn-call pager2FA inboxOutbound follow-upAutonomous workerClaude CodeCursorSupport lineAppointment bookingAI ReceptionistCoding-agent phoneOn-call pager2FA inboxOutbound follow-upAutonomous workerClaude CodeCursorSupport lineAppointment bookingAI ReceptionistCoding-agent phoneOn-call pager2FA inboxOutbound follow-upAutonomous workerClaude CodeCursorSupport lineAppointment booking

The flow · end-to-end

Phone → Speech → your endpoint.

Unpod handles the phone and the speech. You receive text at your endpoint and reply with text - the prompt, tools and customer data stay on your side.

Customer

Caller

+91 9XXXX XXXXX

Voice WhatsApp SMS

Dials in, or receives an outbound dial / WhatsApp from your agent.

inbound audio

▶

outbound audio

◀

Unpod · Telecom + Speech

+91 80 4555 0142

STT

streaming · voice → text

~120ms

Route

unified webhook · threads

~30ms

TTS

neural · text → voice

~180ms

All-in60% less cost vs legacy stacks

streamed text

▶

reply text

◀

Your endpoint

Your agent

POST /unpod

✓Your prompt + system

✓Your LLM + tools

✓Your customer data

Stays entirely on your side. We never see the brain.

How it works

Four steps to a working phone agent.

No telecom plumbing. No separate SMS provider. One SDK, one webhook.

Get a number

One API call provisions a real number - voice, SMS and WhatsApp on the same line.

One API call

Point one webhook

Voice and messages arrive in the same event shape. One handler covers both channels.

Single endpoint

Respond with text

Calls are transcribed live and sent as text. Reply with text - we handle TTS, threading, silence.

Text in, text out

Ship it

Your agent is live with a real number. Native MCP in Claude Code, Cursor, Windsurf, Zed.

Live in production

SDKs · open source

Drop it into any backend.

Python, Node, REST or MCP - all open source. Receive text, reply with text. Wire it to any LLM, framework or your own loop.

Read the docs →View on GitHub →

pip install unpod

Python

Node.js

REST

MCP

from unpod import Unpod

client = Unpod(api_key="up_...")

# Provision a number for your agent
number = client.numbers.buy(country="IN", agent_id="agt_123")

# Place an outbound call - your webhook gets every turn
call = client.calls.create(
    agent_id="agt_123",
    to_number="+9197XXXXXXXX",
    webhook="https://api.yourapp.com/unpod",
)

The stack · open core

One stack for voice, messaging, and conversation logic.

SuperDialog writes your conversation logic. The Unpod CPaaS runs the contact-center layer, and you self-host it. The Developer Platform provides the telephony and speech - the only part you pay for.

Speech Infra

Developer Platform

Telephony + Speech

The metered speech & telephony your agents run on. STT, TTS, numbers and routing through one webhook.

Numbers · PSTN · SIP · WebRTC
Streaming STT + TTS
One unified webhook
Metered per-minute

Self-host the full agents & contact-center platform. Embeds SuperDialog, bills against Unpod infra.

Self-host · Docker / Helm
Embeds SuperDialog
Contact-center layer
Apache 2.0

unpod.ai/open-source-cpaasExplore →

Open source

SuperDialog

Conversation framework

Write realtime conversation logic. Standalone, lower-level, free - adopt it before you ever need the cloud.

Realtime dialogue state
BYO LLM + tools
Standalone, no cloud
Free, OSS

unpod.ai/superdialogExplore →

SuperDialogwrites logic for →Unpod CPaaSself-hosts on →Developer Platform= one webhook to your agent

Ready-to-deploy playbooks

What developers build with it

Real things developers ship with Unpod - from coding agents that take calls to on-call pagers and 2FA inboxes.

Phone-call your codebase

Point a number at your server. Call it, say "restart the worker" - your coding agent does it via MCP.

Native MCP

Autonomous AI workers

Give a long-running agent a real number so it can dial vendors and follow up - no human in the loop.

< 10 lines

AI receptionist

Answer inbound 24/7, qualify intent, book meetings, warm-transfer with the full transcript.

1 webhook

On-call & incidents

Pager-style escalation: dial the on-call engineer, read the incident, escalate if no answer.

WSS streamed

Outbound follow-up

Trigger an outbound dial via one API call. Real-time transcript streams back to your handler.

1 API call

2FA & verification

A dedicated number for codes from Stripe, GitHub, banks. SMS arrives as text in the same webhook.

Same schema

Customer support

Handle inbound voice and SMS through one handler. Resolve routine queries, escalate with context.

Voice + SMS

Appointment booking

Outbound dials to confirm bookings; reschedule links go over SMS through the same SDK.

No lock-in

Where we draw the line

We handle voice. You keep your agent.

We handle the phone calls and the speech. You keep your prompt, your model, and your customer data.

CompareTwilioLiveKit / PipeCatUnpod

You send / receiveTwiML + audioAudio in, audio outText in, text out

STT + TTSYou wire itYou wire itOwned by Unpod

Speech quality + voicesYour problemYour problemCurated profiles

Voice + SMS + WhatsAppSeparate APIsVoice onlyOne endpoint

Self-host the platformNoPartialYes · Apache 2.0

5M+

calls handled

across the network

+91

numbers, instant

all major circles + WhatsApp

< 330ms

first-token latency

streaming STT + TTS

99.95%

uptime, regulated-grade

carrier failover built-in

Security & compliance

Enterprise-grade by default.

Audio is streamed, not stored. Data stays in your region, encrypted with your keys - and the open-source layers let you self-host the whole thing.

Data residency

Pin storage to US, EU or India regions - your choice, per project.

Encrypted end-to-end

TLS in transit, AES-256 at rest. Bring your own KMS keys.

Zero-retention audio

Streams aren’t stored by default. Opt in only if you need recordings.

Never trained on

Transcripts and recordings are never used to train any model.

ISO

27001

ISO 27001

Information security

ISO

42001

ISO 42001

AI management

DPDP

India

DPDP India

Data protection

SOC 2

Type II

SOC 2 Type II

In progress

Questions before you start.

The products, the model, integration and security - the things developers ask before shipping.

SuperDialog is an open-source framework for writing realtime conversation logic - standalone and free. Unpod is the open-source CPaaS that embeds SuperDialog and adds the contact-center layer; you self-host it. The Developer Platform (unpod.dev) is the metered telephony + speech infrastructure those layers bill against - the only paid surface.

Build the agent. Ship the voice. Own the stack.

Start free with the open-source framework and CPaaS. Meter against Unpod infra only when you go live. No credit card to start.

Start building →View on GitHub ↗

✓ Free OSS framework✓ Self-host the CPaaS✓ Metered infra only✓ Native MCP

Telephony & Speech Infra for AI Agents.

Telephony and speech delivered as infra - with an open-source CPaaS and a conversation framework on top. Build the agent, self-host the platform, or run the whole stack on Unpod.

One platform · three layers · text in, text out

from unpod import Unpod client = Unpod(api_key="up_...") # Provision a number for your agent number = client.numbers.buy(country="IN", agent_id="agt_123") # Place an outbound call - your webhook gets every turn call = client.calls.create( agent_id="agt_123", to_number="+9197XXXXXXXX", webhook="https://api.yourapp.com/unpod", )