How to Set Up NOVA: A Beginner's Guide

You bought a voice AI that runs on your own machine.

Now you want it talking back.

This is the part people brace for: config screens, key fields, the fear of breaking something.

Forget that. Here is how to set up NOVA (Native Omnimodal Voice Agent): four tabs, a handful of keys, one Save button. If you can paste a password, you can do this.

This is the full beginner walkthrough, start to first conversation, no jargon left standing.

Before You Start: What You Actually Need

NOVA is BYOK. Bring Your Own Keys.

NOVA is the engine. You plug in the brain.

The brain is an AI model from a provider you already trust, reached with a key only you hold.

You do not need every key at once. You need one model key to start talking. Everything else is opt-in, added later, when you want it.

The smallest path to "it works":

• One model provider key (or a local model running on your machine, no key needed).
• Open Settings.
• Pick the model.
• Save.
• Talk.

That is the whole spine. The tabs below give you room to grow into the rest.

Step 1: Open Settings

Launch NOVA. Find Settings.

This is the cockpit. Everything you configure lives here, sorted into a tabbed panel so you are never staring at one giant wall of fields.

The tabs: LLM, Embeddings, Runtime, and Comms.

You will spend most of your first session on exactly one of them.

Step 2: The Four Tabs, In Plain Words

Here is what each tab is for, so nothing feels like a mystery box.

LLM: the brain

This is the one that matters on day one.

The LLM tab is where you choose which AI model thinks for NOVA. You pick a provider, choose a model, and paste a key if that provider needs one.

Providers include the big hosted ones and fully local options that run on your own hardware: Ollama, LM Studio, llama.cpp, and vLLM all work.

If you want to keep everything on your machine, NOVA can detect a local model you are running and use that instead. We cover that path in How to Use Local AI Models With NOVA.

Embeddings: the memory sense

Embeddings are how NOVA turns your notes and files into something it can search by meaning, not just by exact words.

You configure the embeddings provider here.

For a first conversation, leave it. It earns its place when you want NOVA to remember and retrieve, not just chat.

Runtime: the voice

Runtime is the speaking and listening layer: the speech-to-text that hears you, the text-to-speech that answers out loud.

This is what makes NOVA a voice agent and not a chat box.

Glance here to confirm your voice settings before you start talking.

Comms: the reach

Comms is where NOVA learns to talk through other channels: Telegram, Slack, WhatsApp, and Twilio for phone.

You only touch this when you want NOVA reachable beyond your desktop.

Want it answering on Slack or placing real phone calls? Start here, then see Connect NOVA to Slack, Telegram, and WhatsApp.

For your first run, you can ignore Comms entirely.

Step 3: Fill In Your Keys (And the Trick That Makes It Easy)

A key is a long string from a provider that proves the request is yours. You paste it into the matching field. NOVA uses it to reach that service on your behalf.

Here is the part that takes the fear out of it.

Every field has a small info icon next to it. The (i) you can mouse over.

Hover it and you get two things:

• A plain-language explanation of what that field is and why it exists.
• A link that takes you straight to where you get that key.

You are never guessing. You never have to go hunting through a provider's website wondering which page holds the key. The field tells you what it wants and hands you the door to go get it.

The flow is simple. Hover the (i). Read the one-liner. Click the link. Copy your key. Paste it back. Move to the next field.

And remember: you only fill the keys you actually need right now. One model key gets you talking. The rest can wait.

Step 4: Pick a Model

Back to the LLM tab.

Choose a provider, then choose a model from the list.

Running something locally? NOVA can scan for local models and show you what it finds, so you select it from a list instead of typing anything cryptic.

There is no perfect first choice. Pick a capable model from a provider you have a key for.

You can change it later in seconds. NOVA hot-swaps the provider live, so switching models is not a reinstall; it is a click.

Step 5: Hit Save

One button. Save.

Model and provider changes apply live. No restart ritual, no praying it took.

If something is missing, this is where you find out, not three steps later. Fix the field, Save again, done.

Step 6: Have Your First Conversation Out Loud

This is the moment the whole setup was for.

Speak to it. Ask it something real.

"What can you do?" is a fine opener. So is asking it the time, or to open a website, or to summarize what is on your screen.

It hears you through the voice layer you set in Runtime. It thinks with the model you chose in LLM. It answers out loud.

That is NOVA running. On your machine. Owned by you, not rented by the month.

When you want to push it further, NOVA ships with a deep toolset out of the box: over 60 built-in tools on Windows, from web search to desktop control to placing phone calls. See 12 Cool Things You Can Do With NOVA for where to point it next.

If You Get Stuck

Almost every first-run snag is one of three things:

• A key in the wrong field. Re-hover the (i), confirm the field, re-paste.
• No model selected. Go back to the LLM tab and pick one from the list.
• Voice not responding. Check the Runtime tab and confirm your speech settings.

None of these break anything. You hover, you fix, you Save. That is the entire loop.

What You Just Set Up

You opened Settings. You read four tabs in plain words. You pasted the keys you needed, guided by the (i) on every field. You picked a model. You hit Save. You talked, and it talked back.

That is a real, local-first voice agent, configured by you, in one sitting.

From here it only gets bigger. Connect it to your channels. Give it memory. Hand it your local model.

Each piece is the same loop you just learned: open the tab, hover the (i), fill what you need, Save.

You own the whole thing. One purchase, all future updates included. Get NOVA if you have not yet, then point it at something that matters.

Welcome to the cockpit.