# Core concepts

> The handful of terms that explain how Lorito fits together.

Source: https://www.lorito.ai/docs/getting-started/core-concepts

Lorito has a small set of building blocks. Once you know these, the rest of the product makes sense. The big picture is **project → agent → channels**: a project holds your knowledge, an agent answers from it, and channels are where that agent appears.

## Project (knowledge base)

A **project** is your knowledge base and the home for everything related to one site or client. It holds your content, your agent, and your channels in one place. Agencies typically run one project per client.

## Knowledge source and items

A **knowledge source** is something you give Lorito to learn from, such as your website, a PDF, a block of text, or markdown. A single website source contains many **items**: the individual pieces of content Lorito has read and can answer from. You can add and remove sources at any time, and Lorito keeps your index in step.

See [Adding knowledge sources](https://www.lorito.ai/docs/knowledge-base/adding-sources) for the full list of source types and how to add them.

## Indexing, embeddings, and retrieval

When you add a source, Lorito reads it and **indexes** it, breaking it into searchable pieces and creating **embeddings**, a numeric representation of each piece's meaning. This is how Lorito understands your content; it does not change the AI model itself.

When a visitor asks a question, the agent **retrieves** the most relevant pieces from your knowledge base and answers using them. This approach is often called RAG (retrieval-augmented generation), and it's what keeps answers grounded in your own content rather than guesses.

## Agent (persona, instructions, model)

An **agent** is the assistant that talks to your visitors. It has a few parts you can shape:

* **Persona.** The name and voice your agent presents to visitors.
* **System prompt.** The instructions that guide how it answers. Lorito drafts these from your content when you set up, and you can edit them any time.
* **Model.** The underlying AI that powers responses. Lorito uses a sensible default, which you can change in agent settings.

Each project starts with one default agent, ready to go.

## Channels and deployments

**Channels** are the surfaces where your agent appears. A **deployment** is one live instance of your agent on a particular channel. Lorito's web channels available today include:

* **Web widget.** A floating chat bubble you add with a `<script>` snippet.
* **Inline embed.** The chat placed directly inside a page.
* **Hosted page.** A ready-made chat page on a Lorito link, with the option to use your own domain.

You set up your knowledge once and can run your agent on any of these. (Channels like WhatsApp and email are coming soon.) Learn more in the [Channels overview](https://www.lorito.ai/docs/channels/overview).

## Conversations

A **conversation** is a single visitor's chat with your agent, the back-and-forth of questions and answers. Lorito records conversations so you can review what people are asking and how your agent responded, which is the best way to spot gaps in your knowledge base.

## Credits

Each answer your agent generates uses **credits**. Your plan includes an allowance, and your current usage and limits are always shown in-product. When you need more capacity, you can change your plan from your account settings.

## Putting it together

You add **sources** to a **project**, Lorito **indexes** them, your **agent** answers visitor questions by **retrieving** from that knowledge, and you deploy it to one or more **channels**, all while keeping an eye on **conversations** and **credits**.
