Advanced

In this tab you can manually edit the RAG Settings for the Resources of your AI.

Resource uploading parameters

Chunk Size

When you upload a resource, its content is split into chunks.

If you upload a resource with the No Split chunking strategy, the whole file is treated as one chunk. This can fail if that chunk exceeds the input length limit of 2000 tokens.

• Max value: 200 – 2000 tokens

• Defines the maximum size of a single chunk of text.

Chunk Overlap

Chunk overlap means that consecutive text chunks share some of the same content, so important information that sits at the boundary between chunks is not lost.

• Range: 0% – 25% of the chunk size

• Controls how much of the end of one chunk overlaps with the beginning of the next. Increase above 0% if you want content continuity, up to a maximum of 25%.

Chunk 1: "The company was founded in 2010 by John Smith." Chunk 2: "John Smith. It expanded to Europe in 2015."

In this case, "John Smith." is the chunk overlap between the first and the second chunk

With a chunk size of 1000 and 25% overlap, each chunk still contains a total of 1000 tokens of which 250 tokens are the repeated ones the adjacent chunk.

Improving retrieval of resources for the AI reply

Limit defines how many candidate chunks are retrieved.

Score Threshold removes candidates whose relevance score is below a minimum value.

Limit

• Minimum: 100 chunks

• Specifies the number of chunks the AI will consider when generating an answer.

Score Threshold

Score Threshold is the minimum relevance score a chunk must have to be included in the results.

• Range: 0.3 – 0.9

• After the system fetches the chunks ( up to the Limit ), each chunk is assigned a similarity score based on how closely it matches the query.

• Lower values (e.g., 0.3) allow more chunks to be considered, while higher values (e.g., 0.9) are stricter, only selecting highly relevant chunks.