Settings Overview

All settings are adjustable from the Settings page in Locke.

AI Models

LLM Model

The chat model used for answering questions. Default: llama3.1.

Supported models include Llama 3.1, Mistral, Gemma, and any other model available through Ollama.

Embedding Model

The model used for generating semantic search vectors. Default: nomic-embed-text.

Changing the embedding model will require re-indexing all documents.

Vision Model

An optional model for extracting text from images and photos. When enabled, Locke can index screenshots, scanned documents, diagrams, and other image files alongside your text documents.

Chunking Settings

Chunk Size

Number of characters per document chunk. Default: 500.

Smaller chunks give more precise retrieval but may lose context. Larger chunks preserve context but may include irrelevant content.

Chunk Overlap

Number of overlapping characters between adjacent chunks. Default: 50.

Overlap ensures that information spanning chunk boundaries is not lost.

Retrieval Settings

Top K

Number of chunks retrieved per query. Default: 10.

Higher values give the LLM more context but may introduce noise. Lower values are faster and more focused.

Temperature

Controls model creativity. Default: 0.7.

0 — focused, deterministic responses
0.7 — balanced (recommended)
2 — highly creative, less predictable

System Prompt

You can customize the system prompt to change how Locke responds. The default prompt instructs the AI to answer based on the provided document context and cite sources.

General Chat

When Allow General Chat is enabled (default: off), Locke will respond using its own knowledge even when no documents match the query. When disabled, Locke only answers questions it can ground in your documents.

Appearance

Locke ships with 4 color themes: Cyan, Violet, Rose, and Amber. Switch themes from the Settings page.