Getting started

Your assistant runs on its own private instance and talks to you through a

messaging app you already use. Getting one running comes down to three

decisions — this page walks you through them in the order the deploy

wizard asks.

(New here? Your account covers signing in and where

your data goes. If a term on this page is unfamiliar,

Words we use is the short glossary.)

1. Pick a plan

The plan is the server your assistant runs on. There are three —

Billing & plans compares what each

includes and links to current prices. Minimal is enough for one

assistant doing everyday work.

2. Pick a channel

The channel is where you'll chat with your assistant:

  • Telegram — the fastest way to start. We can

create a bot for you automatically during deploy, or you can bring your own.

<!-- Withdrawn 2026-08-19 (restore with ChannelsTab.jsx and docsCatalog.js):

  • Discord — bring your own Discord bot and

invite it to a server. Set up after deploy, from your dashboard.

  • WhatsApp — links to a WhatsApp number by QR

code. Read the risk notes first; set up after deploy, from your dashboard.

-->

You pick your channel during the deploy wizard.

3. Choose what powers the AI

The assistant thinks with an AI model. The wizard gives you two ways to

supply one, and leads with the one that needs no setup at all.

Niva AI credits — the default

Niva AI credits comes preselected: we supply the AI. There's nothing

to sign up for and no key to paste. You choose how much AI credit you

want each month — it's required (e.g. R 100, R 250 or R 500 per month) —

and it's added to your plan, charged as one payment on one invoice.

Credits deployments bill monthly only; annual billing isn't available

with credits.

You can't overspend. The cap is enforced by the AI provider itself, on a

key that belongs only to your assistant, and there is no automatic

recharge — ever. If credit runs out, your assistant stops replying;

your instance keeps running, and topping up brings it straight back.

Top-ups, rollover and the low-credit warnings are covered in

Niva AI credits.

Bringing your own provider

If you already have an account with an AI company and want to use it,

choose "I'd rather use my own provider" on the same step. You bring

either an API key (pay-per-use — that provider bills your AI usage

separately) or, for some providers, an existing subscription

(ChatGPT or Claude). See

Choosing a provider for a comparison, and

the per-provider guides for exactly where to create a key:

OpenAI ·

Claude (Anthropic) ·

Google Gemini ·

OpenRouter ·

DeepSeek ·

Moonshot AI ·

OpenCode Zen

Your credential is used only to configure your own instance. Treat every

key like a password: don't share it, and rotate it at the provider if you

think it leaked.

Deploy and say hello

The wizard runs in this order: plan, a name for your assistant, channel,

AI, model, then review, pay and deploy.

On the model step a recommended model is already selected. With credits,

keeping it is the normal choice — after deploy you don't deal in model

names at all, you pick between three options (Everyday, Smarter

or Most capable) on your dashboard's Providers tab. With your own

key, the list starts as that provider's common models and updates to the

ones your account can actually use once your key is checked.

When the instance is up you'll get a QR code or link to open your bot —

send it a message and you're talking to your assistant.

After you're running

dashboard, statuses, and the difference between Restart and Stop.

provider any time, no redeploy needed.

When something doesn't work

Start with Troubleshooting — most issues

are on that page. For everything else,

Support, bugs & feature requests covers how

to reach us and what details make a report easy to act on.