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

# Chunking

> How Jabrod splits your documents

# Chunking

LLMs have limited context windows, and vector databases perform best on smaller, focused pieces of text. **Chunking** is the process of splitting large documents into smaller pieces (chunks) before embedding them.

Jabrod supports several chunking strategies that you can configure per-pipeline.

## Strategies

### Fixed Size

*(Available on Free and Pro)*

Splits text into chunks of an exact character length, with a slight overlap to prevent cutting context in half.

* **Pros:** Fast, predictable, works well for unstructured data.
* **Cons:** Can cut sentences or thoughts in half.

### Sentence-Based

*(Pro only)*

Splits text at natural sentence boundaries (periods, exclamation marks).

* **Pros:** Keeps complete thoughts intact.
* **Cons:** Can result in very short or varying chunk sizes.

### Paragraph-Based

*(Pro only)*

Splits text at double newlines.

* **Pros:** Excellent for structured documents and articles.
* **Cons:** Can fail if the document has erratic formatting.

### Recursive Character

*(Pro only)*

Tries to split by paragraphs, then falls back to sentences, then to words if a chunk is still too large.

* **Pros:** The most balanced approach for general-purpose documents.

### Semantic Chunking

*(Pro only)*

Splits text into sentences, embeds each sentence, and groups them together as long as their cosine similarity stays above a dynamic threshold. When a topic shift is detected, a new chunk is started.

* **Pros:** Extremely high quality. Keeps topically coherent information together.
* **Cons:** Slower to process because it requires embedding every single sentence during ingestion.

## Overlap

Most strategies allow you to configure an **Overlap**. This means the end of Chunk A is repeated at the beginning of Chunk B. This ensures that if a search term crosses a chunk boundary, context isn't lost.