Quick start

This guide takes you from a fresh install to your first AI conversation.

Before you begin

You need one of the following:

• A working Ollama install on your machine (easiest, fully local), or
• An API key for any one of: OpenAI, Anthropic, Google Gemini, OpenRouter, DeepSeek, or
• The Local AI edition of Verzeta Studio, which includes the on-device inference engine (no server and no account: download a model in the wizard and run everything locally).

You do not need all of them. You can add more providers later.

Which download did I get?

Each release ships two editions per platform:

Step 1: Launch the app

Open Verzeta Studio. On first launch you will see the Get Started wizard.

Note: If you skip the wizard, you can return to it from Settings → Onboarding → Re-run Welcome Wizard at any time.

Resizing the window

The window adapts as you make it narrower, so the app is usable on a small screen or a phone-sized window as well as a full desktop one.

• Wide: the navigation bar, the conversation list and the settings panel all sit side by side.
• Narrower: the navigation bar and conversation list move into a drawer you open with the menu button in the top left, and the canvas gets its own tab instead of sharing the screen with the chat.
• Phone width: cards stack their controls, tab labels shorten, and secondary tags on provider and agent cards are hidden so names and descriptions keep the room they need. Everything hidden this way is still available once you widen the window again.

Step 2: Configure one provider

In the wizard's Providers step, you will see these sections:

• Cloud Providers: OpenAI, Anthropic, Google Gemini, OpenRouter, DeepSeek
• Remote Servers: Ollama, llama.cpp (Remote)
• Local Models: the bundled local inference engine (Local AI edition)
• Custom OpenAI-Compatible Servers: any other server that speaks the OpenAI API (see Self-hosted OpenAI-compatible servers)

Pick one to start and click its card.

• For Ollama: leave the base URL at http://localhost:11434 (the default). Click Test Connection. You should see a green check.
• For a cloud provider: paste your API key. Click Save.
• For llama.cpp (Remote): enter the URL of your running llama.cpp server (default http://localhost:8080/v1).

Pro-tip: You can configure more providers later from Settings → Text Providers. The wizard only needs one to let you continue.

Step 3: Local Intelligence (optional, skippable)

The wizard's Local Intelligence step sets up two optional on-device models. Both are only offered in the Local AI edition; in the Standard edition the step points you to your remote providers.

• Routing model: decides which agent replies next in group chats. Choose Use the on-device model and either pick a .gguf file you already have (Browse…) or click Download recommended (~2.3 GB, verified against a pinned checksum). Runs on your GPU via Vulkan when available, otherwise on CPU.
• Embedding model: powers RAG retrieval from your messages and documents. Same idea: Browse… a .gguf or Download recommended (~139 MB).

Nothing downloads unless you click the button, and no models are ever bundled with the app. Downloads come from Hugging Face.

Note: You can change all of this later in Settings → Providers → RAGP (Routing & Gating) and Settings → Providers → Embeddings (RAG). Both pages have the same download button.

If you are only using solo chat with a remote provider, skip this step entirely.

Image generation and web search (optional)

The next two wizard steps show the same pages as Settings:

• Image Generation: add an image provider and set it active so agents and the Generate Image button can create images.
• Web Search: agents search with DuckDuckGo by default. Pick another provider (and enter its key) for better results, and use Test to check it works.

Both can be skipped and changed later in Settings → Providers → Image Generation and Settings → Providers → Web Search.

Step 4: Send your first message

Click Done to finish the wizard.

On the home screen, click New Chat.

In the input field at the bottom, type a message such as:

Tell me one interesting fact about the Voyager 1 spacecraft.

Press Enter to send. (Shift+Enter for a newline.)

The reply appears as it is generated. While it streams, the Send button turns into Stop, which cancels the reply. Afterwards, Retry last response in the message bar sends your last message again for a new reply.

Step 5: Try a group chat

On the home screen, click New Group Chat.

In the New Group Chat dialog:

• Enter a Title.
• Under Members, click Add Member, pick an agent template (e.g. Researcher, Code Reviewer, Designer) and give it an alias. Add at least two members.
• For each member, you may optionally press the configure button on its row (Configure model / tools for this member) to set a different Provider, Model or tool whitelist.
• Tick Mark as coordinator on one member (suggested: the first one).
• Click Create.

Send a message addressed to the coordinator, such as:

@Alice, research two Python CLI libraries and tell me which to use.
Cite real sources.

Watch the conversation cascade. Agents @mention each other, hand off work, and the coordinator drives the team.

Note: Each round allows 10 agent turns and 3 turns per member. While the team makes progress it starts new rounds on its own, up to 6 by default. When it stops, the chat says why. Reply with anything to start it again.

What's next

• Open Chat Settings in a conversation's header to change the model, system prompt, or tools for that one chat. See Conversations.
• Open the Activity Log for your conversation to see the audit trail. See Activity timeline.
• Pair an Android device. See Remote access and paired clients.
• Install a skill so agents can use specialised knowledge. See Skills.