---
title: "Why do I need to regenerate the knowledge after changing the embedding model?"
description: "Vectors from the old and new embedding models don't share a vector space, so retrieval goes off target until you regenerate; plus the steps to re-sync each file source"
canonical_url: "https://docs.ardge.com/en/knowledge/argent/regenerate-knowledge"
---
# Why do I need to regenerate the knowledge after changing the embedding model?

> Vectors from the old and new embedding models don't share a vector space, so retrieval goes off target until you regenerate; plus the steps to re-sync each file source

## Short answer

The existing knowledge is made of vectors produced by the previous embedding model. They don't share a vector space with vectors from the new model, so the two can't be compared. Without regenerating, the agent's retrieval results go off target. ARGENT doesn't rebuild these vectors automatically — you have to re-run the sync for each file source, one at a time.

## Details

### The notice in the interface

When you change the **Embedding Model** on the **2. Model Settings** tab of the **Model Connection** page, the field shows **Changing the embedding model requires re-generating the AI agent knowledge manually.** below it.

### Why it has to be redone

When you create a file source, ARGENT splits the file contents into chunks, has the **Embedding Model** turn them into vectors, and stores them. When someone asks a question, the question is also turned into a vector first and compared for similarity against the stored vectors. The most relevant chunks then go to the language model to compose the answer.

This comparison only holds when the question and the chunks were converted by the same model. After you change the embedding model, new question vectors are compared against chunk vectors the old model left behind. The similarity scores no longer mean anything, so retrieval goes off target. Re-running the sync converts the files again with the new embedding model, which puts both sides back in the same vector space.

## How to regenerate the knowledge

Do the following for each file source:

1. Go to the **File Sources** page
2. At the right of the target file source's row, click the sync icon
3. The confirmation dialog shows **Start Sync** and **This will start syncing "" files**. Click **Start**

Once it starts, the page shows **Sync job has started**. The row's **Last Run** column switches to a progress percentage, then shows **Completed** when processing is done.

File sources created with **Manual Upload** and with **Select from NAS** can both be re-synced the same way.

<warning>

**Re-sync every file source separately:**

The sync icon only reprocesses that one file source. The other sources in the list still hold vectors from the old model, so you have to go through them one by one.

</warning>

<tip>

**Stop a source that's mid-sync first:**

For a file source whose status is **In Progress**, clicking the sync icon opens the **Stop Sync** confirmation dialog. Stop the current sync, wait for the status to change to **Stopped**, then start it again.

</tip>

## Related articles

- [Set up the model connection in ARGENT](https://docs.ardge.com/en/knowledge/argent/model-connection)
- [What do the file source statuses mean?](https://docs.ardge.com/en/knowledge/argent/source-statuses)


## Sitemap

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