Introduction

Welcome. oyehello lets you build AI voice agents - voice workers that pick up the phone and hold a real, natural conversation with your customers, in the language they actually speak. One agent can call thousands of people to remind them about a payment, or answer every incoming support call at 2am, or qualify a fresh batch of leads before your sales team ever picks up. It sounds human, it never gets tired, and it sticks to the brief you gave it. You build it once, in your browser, with no code.

This guide walks you through the whole thing - from your first agent to running real campaigns and reading the results. Take it at your own pace; nothing you build goes live until you decide it should.

New here? You don’t need to read this end-to-end. Skim this page to get the mental model, then jump to whatever you’re trying to do. The sidebar is grouped the same way you’ll actually work: build an agent → make calls → review results.

The mental model

Almost everything in oyehello hangs off one simple idea:

You build an agent once, then point calls at it - and review what happened afterward.

That’s the whole loop. Three moving parts:

The agent is the “who.” It’s the voice worker that does the talking - its voice, its language, its persona, its goal, the rules it follows, and the things it’s allowed to do on a call. You build and reuse it. See Create an agent.
The calls are the “how they happen.” An agent doesn’t dial anyone by itself. You either point an outbound campaign at it (to make calls to a list) or point a phone number at it (to answer calls that come in). Same agent, two directions.
The results are the “what happened.” Every call leaves behind a recording, a transcript, an outcome, and any details you asked the agent to capture - all waiting for you to review.

If you remember nothing else, remember that shape: build → call → review. The rest of this guide is just the details of each part.

The building blocks, briefly

Here’s how the pieces you’ll meet fit into that loop:

Agents - the voice worker. You give it a voice and language, a persona, a goal, and a personality. This is where most of your time goes, and it’s worth it.
Knowledge base - your real documents, FAQs, and policies, so the agent answers from your actual content instead of guessing.
Tools & actions - things the agent can do mid-call: transfer to a human, book a meeting, send a WhatsApp, push an outcome to your CRM.
Compliance packs - the rulebook the platform enforces before it ever dials: calling-time windows, Do-Not-Disturb scrubbing, how often the same person can be contacted. You pick a pack; the platform obeys it automatically.
Campaigns - upload a list, map your columns, and let the dialer place calls at scale for outbound work.
Phone numbers - the numbers you call from, and the numbers customers call in to for inbound.
Calls & transcripts and dispositions - everything the call left behind, ready to read, filter, and report on.

What people actually use it for

The same build-once-then-call pattern covers a surprising range of jobs. A few common ones, with the shape of a real call:

Sales & lead qualification - call a list of new leads, figure out who’s genuinely interested, and book the hot ones straight into a calendar. “Hi Priya, you enquired about our home loan - mind if I ask two quick questions to see what you’d qualify for?”
Reminders & confirmations - appointment reminders, delivery confirmations, renewal nudges. Often a short Announcement agent: one clear message, a chance to acknowledge, done. “This is a reminder that your appointment is tomorrow at 4pm. Reply if you’d like to reschedule.”
Customer support - answer incoming calls, resolve the easy stuff from your knowledge base, and warm-transfer the rest to a human. “I can see your order shipped this morning - want me to text you the tracking link?”
Collections & payment follow-ups - remind customers a payment is due and offer a link, with the extra legal safeguards a regulated call needs turned on automatically.
Feedback & surveys - call after a purchase or a visit, ask a few questions, and capture the answers as clean, reportable fields.

Don’t see your exact use case? If it’s a phone conversation with a clear purpose, an agent can probably do it. Start with the closest example above and adjust the brief.

Your first 15 minutes

The fastest way to understand oyehello is to build a tiny agent and hear it talk. Here’s a path that gets you there without touching anything you don’t need yet:

Create an agent. Open the agent builder, give it a voice and language, and write a one-line purpose - “Remind the customer their appointment is tomorrow and confirm they’ll attend.” That’s genuinely enough to start. The full walkthrough is in Create an agent.
Test it before anyone real hears it. Run a simulation - the platform role-plays a customer and scores how your agent did - then do a test call to your own phone so you hear it live. Both are covered in Test your agent.
Publish it. Nothing goes live until you hit Publish. Once you do, campaigns and phone numbers can point at it. Change something later? Just republish.
Run a small campaign. Upload a short list - even five of your own numbers - map the columns, and let it dial. Watch the calls land in Calls & transcripts in real time. See Campaigns.

Tip: keep your first list tiny and use numbers you own. Reading your own transcripts back is the single fastest way to learn what to tweak in the brief.

One thing before you go live

You can build and test agents right away - that part’s free and open. To place live calls to real customers, you’ll first do two quick things: complete KYB verification (so we know who’s calling) and top up your prepaid wallet (calls are billed per minute). Both are walked through in Sign up & workspace and Billing & wallet.

And if you’d rather drive all of this from your own code instead of the console, there’s a full API reference that does everything this guide does, programmatically.

Next: Sign up & workspace → - create your workspace, get verified, and add your team.