# Instfy AI: Instfy Agent overview

<p class="lead">Meet Instfy Agent: an AI agent that serves visitors on your website and through a public link, answering based on your company's content.</p>

Instfy AI is available today through the **Instfy Agent** product, at [ai.instfy.com](https://ai.instfy.com). It works like a pre-sales rep and support agent: it answers questions, recommends products with real links and directs the visitor to your WhatsApp or your team. Use it when you want to serve customers at any time without growing your team.

!image[Instfy Agent widget open in the corner of a website, with suggested questions]

## What Instfy Agent is

Instfy Agent is a platform for creating AI agents. Each agent has its own knowledge base, appearance and channels. You configure everything in the **Studio**, the admin panel at `ai.instfy.com/studio`.

The agent:

- Answers in the visitor's language (Portuguese, English or Spanish).
- Uses only the content you added and the context of the page the visitor is on.
- Shows the **sources** used in the answer, so the visitor can check them.
- Asks one question at a time and, when it detects buying intent, asks for name and contact details in the conversation.
- Does not make up prices, deadlines, stock, warranties or commercial terms that are not in your content.

## Available channels

All channels use the same agent and the same knowledge base.

| Channel | How it works |
|---|---|
| Website widget | A code snippet installs the chat window on any website (HTML, WordPress, Shopify, Google Tag Manager, React or Next) |
| Public page | The agent's own link to use in your bio, on WhatsApp or in a QR code |
| Subdomain | An address in the format `youragent.instfy.com`, which opens the public page |
| Kiosk or POS | Full screen for use at a counter, physical store or event |
| Voice | A page where the visitor speaks and hears the answer, in the browser |

The widget also reads the context of the open page, such as title, description and product data published on the website, to answer about the item the visitor is looking at.

:::note
A native mobile app, the customer's own domain (for example, `chat.yourcompany.com`) and real-time (streaming) answers are **On the roadmap**.
:::

## Knowledge sources

In each agent's **Knowledge** tab, you add content in four ways:

| Type | What goes in |
|---|---|
| Page | A specific URL on your website |
| Website | A crawl starting from the main address and the sitemap, from 30 pages (default) up to 120 pages |
| File | TXT, MD, CSV, TSV, JSON, HTML or XML, up to 5 MB |
| Text | Free text or an FAQ written by you |

Each source shows its indexing status and has the **Refresh** (to fetch a URL's content again) and **Remove** buttons.

:::note
Uploading **PDF and DOCX** files is **Coming soon**. In the meantime, convert the content to TXT or paste it as Text.
:::

## How answers are grounded

1. The visitor asks a question.
2. The system searches only the agent's knowledge base for the passages most relevant to the question.
3. The AI model receives those passages, the page context and the agent settings, with the rule to use only that information.
4. The answer comes back with the list of sources used.

If the answer is not in your content, the agent says it does not have that information. These questions appear as **Knowledge gaps** in the Analytics tab, so you can complete the knowledge base.

## Studio areas

- **Dashboard:** an overview of the organization, with conversations, leads, pipeline, channels and most frequent questions.
- **Agents:** the list of agents and, inside each one, the **Builder**, **Knowledge**, **Analytics**, **Conversations**, **CRM** and **Install** tabs.
- **Integrations:** connectors that are **On the roadmap** (see [Management and automation agents](#/ai-agents)).
- **Settings:** organization, team and access levels (Owner, Administrator, Manager, Editor, Client and Viewer).

## Privacy, in short

- The agent accesses **only** its own knowledge base, the page context and the settings. It does not access your ERP, email or other systems.
- Conversations and messages are saved and can be reviewed in the **Conversations** tab, along with the source channel.
- The visitor's browser stores the session and chat history locally, so the conversation continues when they change pages.
- Messages are processed by AI models from providers contracted by Instfy to generate the answer.
- Removing a source deletes its indexed content. Deleting an agent deletes the conversations, leads and events linked to it.
- Indexed content is treated as data, never as instructions, which reduces the risk of third-party text manipulating the agent.

For requests about personal data, write to relationship@instfy.com.

## Frequently asked questions

### Are Instfy Agent and Instfy AI the same thing?

Yes. Instfy AI is Instfy's artificial intelligence area, and the product available today is called Instfy Agent, at ai.instfy.com.

### Can the agent make up information?

The agent is instructed to use only your content and not to make up prices, deadlines, stock or terms. When it cannot find the answer, it says so, and the question appears as a knowledge gap.

### What languages does the agent answer in?

The agent answers in the visitor's language, with support for Portuguese, English and Spanish.

### Does the agent connect to ERP Instfy?

No. Today the agent uses only its own knowledge base. Integrations with other systems are on the roadmap.

### Can I upload PDFs to the knowledge base?

Not yet. PDF and DOCX uploads are coming soon. Use website pages, TXT, MD, CSV, JSON, HTML or XML files, or paste the content as text.

## Next steps

Create your first agent and see how to use the management features.

:::related
ai-setup, ai-agents, ecosystem
:::