## 1. Create Knowledge-Base

First, let's say there're text files explaining ai. We'll build a knowledge-base with from the text files. The directory should look like below.

```
ai_tutorials/
  |
  *-- ai_tutorial_1.txt
  |
  *-- ai_tutorial_2.txt
  |
  *-- ai_tutorial_3.txt
  |
  *-- ... and many more txt files
```

Run `cd ai_tutorial; rag init`. You'll see a new directory created like below.

```
ai_tutorials/
  |
  *-- .ragit/
  |   |
  |   *-- chunks/
  |   |
  |   *-- configs/
  |   |
  |   *-- files/
  |   |
  |   *-- images/
  |   |
  |   *-- prompts/
  |   |
  |   *-- index.json
  |   |
  |   *-- models.json
  |
  *-- ai_tutorial_1.txt
  |
  *-- ai_tutorial_2.txt
  |
  *-- ai_tutorial_3.txt
  |
  *-- ... and many more txt files
```

`.ragit/` is like `.git/` of git repositories. It saves metadata and chunks. After `rag init`, the knowledge-base is empty. You have to add files to the staging using `rag add` command.

Run `rag add --all`. Now you're ready to build the knowledge-base. Run `rag build` to start the work. The default model is `llama3.3-70b-groq` and you need `GROQ_API_KEY` to run. If you want to run gpt-4o-mini, run `rag config --set model gpt-4o-mini`. You can see the list of the models using `rag ls-models`. You can also add models manually to `.ragit/models.json`.

```
elapsed time: 00:33
staged files: 15, processed files: 13
errors: 0
committed chunks: 39
buffered files: 8, buffered chunks: 8
flush count: 1
model: gpt-4o-mini
input tokens: 14081 (0.001$), output tokens: 1327 (0.000$)
```

`rag build` takes very long time and money (if you're using a proprietary api). It creates chunks and add title and summary to each chunk, using AI.

You can press Ctrl+C to pause the process. You can resume from where you left off by running `rag build` again. (more on [a dedicated document](./commands/build.txt))

```
ai_tutorials/
  |
  *-- .ragit/
  |   |
  |   *-- chunks/
  |   |   |
  |   |   *-- ... a lot of directories
  |   |
  |   *-- configs/
  |   |
  |   *-- files/
  |   |
  |   *-- images/
  |   |
  |   *-- prompts/
  |   |
  |   *-- index.json
  |   |
  |   *-- models.json
  |
  *-- ai_tutorial_1.txt
  |
  *-- ai_tutorial_2.txt
  |
  *-- ai_tutorial_3.txt
  |
  *-- ... and many more txt files
```

After it's built, you'll see many data files in the `.ragit/` directory. You can ask queries on the knowledge-base now.

NOTE: You can ask queries on an incomplete knowledge-base, too.

## 2. (Optional) Clone Knowledge-Bases from web

This is the key part. You can download knowledge-bases from the internet and extend your knowledge-base with those. You can also share your knowledge-base with others.

First, let's make a fresh directory. Run `mkdir playground; cd playground`.

```
playground
```

Before downloading knowledge-bases, we have to init a rag index. Run `rag init`.

```
playground
  |
  *-- .ragit/
      |
      *-- chunks/
      |
      *-- configs/
      |
      *-- files/
      |
      *-- prompts/
      |
      *-- index.json
      |
      *-- models.json
```

You'll see an empty rag index. Now we have to download knowledge-bases from the web. I have uploaded a few sample knowledge-bases for you. You can `rag clone` them, like `rag clone http://ragit.baehyunsol.com/sample/git`