Skip to main content
Unpod is the agentic backend for conversational AI. Apps that talk, remember and act: voice, messaging, agents and workflows in one SDK, on a Context Engine that remembers every customer across every channel. Today, voice and the Chat API run in production, and voice agents remember past calls; WhatsApp, SMS, email and cross-channel memory are in build. Build the app. Unpod runs everything behind the conversation. Supabase simplified the backend for traditional apps; Unpod is the backend for apps that communicate, remember and act.

​
A conversational backend is not a normal app backend

Every team building an app that talks assembles three hard layers from separate systems, then runs them for every user.

Channels you don't own

Each channel has its own carrier, policy and approval rules. Call, WhatsApp, SMS and email each have to be built and kept compliant separately.

The agent harness

Running an agent in production means tool calls, guardrails, retries, human handoff and evals - during and after every call.

Context that compounds

One identity and one memory per customer, across every channel and session, ready before the agent speaks.
Then replicate it for every user. A consumer app has millions of users; a SaaS product, thousands of accounts. For a small team, this layer alone is a distributed-systems problem. Unpod is built to be that one system - see what runs today, below.

​
One backend, every layer

Every layer shares one model of the person and the conversation. Start with the person, not the request:
1

A channel event resolves to a person

Someone calls or writes; Unpod finds who they are and what happened before. Calls, same number: production Messages, cross-channel: in build
2

The harness loads context

It picks the agent and its tools and applies the guardrails.
3

The flow advances

An agent - or a human - takes the next step.
4

The outcome is written back

The next interaction starts informed. Next call, same number: production Any channel: in build

​
What runs today

Voice is in production on Unpod’s own rails: agents and the agent harness, per-call customer context with memory of past calls, and a structured outcome after every call. WhatsApp, SMS and email conversations, cross-channel context, and journeys are in build. Every Core Engine page prints its own status.
10+ lakhInteractions / month
~400 msCall latency
99.94%Reliability
MITOpen source

​
What you can build

Apps on Unpod

One backend, many apps: a contact center, a personal assistant, an email-first agent - and our own platform.

Who it's for

Conversation-first startups, SaaS platforms, contact centers and regulated enterprises.

​
Start building

Getting Started

Choose your path - and the right credential for it.
Try itPlaygroundBuild and hear an agent in the browser. No account needed.

GitHub

Open source, MIT. Self-host the platform; voice deploys separately.