> For the complete documentation index, see [llms.txt](https://docs.bonzai.sh/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.bonzai.sh/start-here/first-10-minutes.md).

# Your First 10 Minutes

This walkthrough gets you from installation to a useful result without technical setup.

## Minute 1: choose one simple outcome

Pick something small and real:

* summarize a document;
* draft a customer reply;
* create an image concept;
* capture a useful webpage image;
* ask a companion to remember a preference.

Starting with a real task makes model choices much easier to understand.

## Minutes 2–4: let BonzAI prepare

When BonzAI asks to download a model:

1. Read the model's purpose and hardware estimate.
2. Choose a smaller or faster model for your first attempt.
3. Keep BonzAI open while the progress bar completes.

The download is stored locally. It should not repeat every time you launch the product.

## Minutes 5–7: make your first result

### In BonzAI Web

Type a question and send it. If local browser AI is unavailable on your device, BonzAI explains which product can complete the task.

### In BonzAI+

Open **Chat** and ask anything. To ask about a webpage image or text passage, use the **BonzAI+** capture button on the page, then return to Chat.

### In BonzAI Desktop

Open **Chat**, choose the recommended local model, and describe the outcome you want. You can change models later without losing the purpose of your conversation.

## Minutes 8–10: make the result reusable

BonzAI is most useful when work does not disappear after one answer.

* Keep the conversation if it contains useful context.
* Save a generation to history.
* In BonzAI+, save a captured example in **Dataset** mode.
* Give a companion a fact worth remembering.
* Export or mint only when you deliberately want the work to leave its private state.

## A good first prompt

Use this shape:

> I want to achieve **\[outcome]**. The audience is **\[person]**. Use **\[source/context]**. The result should feel **\[tone]** and must include **\[important constraints]**.

Example:

> I want to reply to a late-paying customer. The audience is a long-term client. Be warm but clear, mention invoice 1042, and ask for payment by Friday without sounding threatening.

## If the result is slow

* Pick a smaller model.
* Close other memory-heavy applications.
* Reduce image or video resolution.
* Use a faster generation mode.
* Let an active generation finish before starting another large one.

Slower hardware does not prevent you from using BonzAI. It changes which model and quality setting will feel comfortable.