Knowledge Lab — overview and models

Knowledge Lab takes facts out of your database and makes a local language model able to recall them. Where the AI assistant sends schema and a handful of sample rows with each question, Knowledge Lab prepares the knowledge once, up front, from a SQL query.

Open it from Tools → Knowledge Lab (LLM knowledge).

Requirements

Knowledge Lab is included with every license. It additionally needs the LARQL add-on, which is a separate binary:

<app-data>/plugins/larql/larql.exe

On Windows that is %APPDATA%\com.nabusql.app\plugins\larql\larql.exe. Copy the file there and restart NabuSQL. Until you do, the tool opens and tells you the add-on is missing — every command is a no-op rather than an error.

The tabs

Tab	Purpose
Models	Prepare and manage local models
LQL console	Query a model directly
Knowledge test	Check what a model recalls
Patches	Manage stored knowledge and export it

Preparing a model

Add GGUF model and pick a .gguf file. NabuSQL converts it into a vindex — the internal query format — automatically.

The conversion happens once per model. It uses extra disk space and leaves the original file untouched, so nothing you already use with another tool is affected.

For models you already have in another form, Advanced: folder (safetensors/vindex) takes a directory instead.

Prepared models are listed with a capability:

• inference — the model can generate. The full feature set is available.
• browse only — the vindex can be inspected and queried, but not used for generation.

Delete model removes the model and its vindex from disk, after confirmation.

Where things are stored

<app-data>/knowledge/models/<name>.vindex/     prepared models
<app-data>/knowledge/patches/<model>/<name>.vlp knowledge patches

Models are large. Check the free space on that drive before preparing several.

Next

• From SQL into a model — the wizard that turns query results into knowledge.
• LQL console and patches — querying, testing and exporting.