---
title: "ARGENT"
description: "Build an AI agent that answers from your own documents — set up the models, add file sources, create the agent, then publish it and watch how it's used"
canonical_url: "https://docs.ardge.com/en/guide/ai-applications/argent"
---
# ARGENT

> Build an AI agent that answers from your own documents — set up the models, add file sources, create the agent, then publish it and watch how it's used

## Overview

ARGENT turns your company's documents into knowledge an AI can search, then builds an AI agent that answers questions based on that content.

For example, you can hand ARGENT your product manuals, SOPs, contracts, FAQs or the shared documents on your NAS, and people can then ask it questions directly. ARGENT first finds the relevant content in the documents you specified, then has the AI compose the answer.

**Documents / NAS → build searchable knowledge → create an agent → test the answers → publish**

Everything runs on the ARX device. The models ARGENT relies on are deployed in [AI Model Center](https://docs.ardge.com/en/guide/use-ai/ai-model-center) on the same machine, so your documents never leave the device.

![The ARGENT dashboard](https://docs.ardge.com/screenshots/guide/en/argent.webp)

![A diagram from documents to AI agent](https://docs.ardge.com/screenshots/guide/en/arg-01.webp)

## Before you begin

ARGENT doesn't include every AI model it needs. Before you start, confirm the models ARGENT requires are deployed in AI Model Center: a language model and an embedding model. To process PDF, Office or image files, you can also deploy the OCR and Enhance models to improve text recognition quality.

If you're not sure which to pick, go with the recommended model set App Market offers.

## Step 1: Set up the model connection

On ARGENT's **Model Connection** page, confirm the model sources are reachable, then choose a model already running in AI Model Center for each of the four uses: language, embedding, OCR and Enhance. Nothing else works until this is right.

For the detailed steps, see [Set up the model connection in ARGENT](https://docs.ardge.com/en/knowledge/argent/model-connection).

## Step 2: Decide where your data comes from

- **Manual Upload** — good for a first test, a small number of documents, or building a quick proof of concept
- **Select from NAS** — good when your company's documents already live on a NAS, keep being updated, and you want them synced automatically on a schedule

For the trade-off between the two, see [Should I use Manual Upload or Select from NAS for a file source?](https://docs.ardge.com/en/knowledge/argent/local-or-remote).

<note>

Before you upload anything, check [what file formats and limits ARGENT supports](https://docs.ardge.com/en/knowledge/argent/formats-and-limits) — it saves a round of failed conversions.

</note>

## Step 3: Create your first AI agent

You'll usually set the agent's name, description, prompt and the file sources it uses.

Test with a small, representative set of documents first, confirm the answers head in the right direction, and only then widen the data.

For the full flow, see [Create your first AI agent](https://docs.ardge.com/en/knowledge/argent/first-ai-agent).

## Step 4: Check that it really found your data

Don't just judge how fluent the answer reads. Test whether:

- The question actually found the relevant documents
- The answer is based on your company's data rather than just the model's general knowledge
- The document sources it found are the right ones
- An updated document was re-synced

On the agent's **File Source** tab, use **Search file contents...** to check the retrieval results, then go to the **Settings** tab and ask real questions in the **Test chat**.

## Step 5: Connect a NAS (when you need it)

Point ARGENT at a CIFS/SMB folder instead of uploading files one by one by hand. With a sync schedule set, each sync automatically processes the files that changed.

For how to set it up, see [Connect a NAS shared folder as a file source](https://docs.ardge.com/en/knowledge/argent/nas-file-source).

## Step 6: Publish the agent

Once the answers are good enough, turn on public access, adjust the chat window's appearance, and paste the embed code into a page.

For the steps, see [Embed an AI agent on your website](https://docs.ardge.com/en/knowledge/argent/embed-on-website).

## Step 7: Watch how it's used

See how many questions all your agents have taken, how many tokens they've used, and the content of every conversation — see [View ARGENT usage and chat history](https://docs.ardge.com/en/knowledge/argent/usage-analytics).

## When something doesn't look right, check these first

- Is the model still running?
- Did the file source sync successfully? — [What do the file source statuses mean?](https://docs.ardge.com/en/knowledge/argent/source-statuses)
- Is the file format supported? — [What to do when a file fails to convert](https://docs.ardge.com/en/knowledge/argent/conversion-failures)
- Did retrieval find the right content?
- Do you need to regenerate the knowledge after changing the embedding model? — [Why do I need to regenerate the knowledge after changing the embedding model?](https://docs.ardge.com/en/knowledge/argent/regenerate-knowledge)


## Sitemap

See the full [sitemap](https://docs.ardge.com/sitemap.md) for all pages.