> ## Documentation Index
> Fetch the complete documentation index at: https://docs.allcodex.allmaker.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Upgrading AllCodex

> Keep self-hosted AllCodex updated by pulling new Docker images, recreating containers, and migrating SQLite, vector index, and PostgreSQL data safely.

AllCodex updates include bug fixes, new features, and optimized parser prompts. Upgrading the self-hosted stack takes three steps.

***

## Standard Docker upgrade

AllCodex persists campaign lore, SQLite files, and vector indices in `/srv/allcodex` on your host machine. This allows you to update containers without data loss.

### 1. Pull the latest images

Pull the updated container tags from GitHub Packages:

```bash theme={null}
docker compose pull
```

### 2. Recreate the containers

Recreate and start the services. Docker replaces containers that have new images:

```bash theme={null}
docker compose up -d
```

### 3. Clean up old images

Remove old, dangling Docker images to free up disk space:

```bash theme={null}
docker image prune -f
```

***

## Upgrading database schema and migrations

### AllCodex Core (SQLite)

Core applies schema migrations on startup. The Express server checks the SQLite schema, runs pending SQL scripts, and updates the database.

### AllKnower (PostgreSQL)

If you run AllKnower via Docker, the container entrypoint runs schema migrations (`prisma migrate deploy`) before launching the web server.

If you run a **local development** setup (Method 2), apply Prisma migrations after checking out the latest code:

```bash theme={null}
cd allknower
bun db:migrate
```

***

## Re-indexing vector embeddings (RAG)

If a new AllCodex release changes embedding models or the RAG schema (such as updating Qwen embeddings), re-index your LanceDB vector database:

1. Open the **Portal Settings** tab.
2. Under the **Search Status** card, look for the **Reindex RAG** section.
3. Click **Reindex All Lore** to queue a background job. AllKnower reads all notes tagged with `#lore`, regenerates embeddings, and rebuilds the vector database.
4. Alternatively, click **Wipe DB Lore & RAG** to wipe the LanceDB and Prisma metadata backends.