Resources

The Resources page allows you to upload external documents or add link to a webpage or a file that will be used to train your LLM model.

Uploaded resources are used to power the RAG ( Retrieval-Augmented Generation ) system, which selects the most relevant documents to provide the AI with the context it needs to generate accurate and insightful answers.

The RAG settings are editable in the Advanced page.

Here, you can upload documents that will be automatically indexed and used by your model to answer specific questions.

The supported file extensions are: excel, word doc, pdf, html and txt

Instead of uploading a file, you can add a link to a webpage and we will fetch the revelant information from the HTML of that page, or download the file at the link ( if that's one of the supported ones )

FAQ, guidelines and general documentation are all perfect candidates for resource training.

Chunking Strategies

The content of the file will be split into chunks, in order to improve the AI answers ( after selecting a file to upload, or a link to add ) you will be asked to select the Chunking Strategy.

• No Split

  This is the simplest of them all, the entire content will be placed in a single chunk.

Please be aware that the maximum size is 32k tokens ≈ 128k characters ≈ 60 pages ≈ 125 KB ( of a simple .txt file )

• Recursive Text Splitting

  Using the parameters set in the Advanced page, the content of the file will be splitted in different chunks ( Check the Advanced page for more info regarding this )

• Document Structure Based

  If the document is among the required ones, the content will be splitted in chunks using a smart algorithm. ( For example, for pdf, the document might be splitted into its chapters, if present, or other way )

Embedding Status

Documents uploaded here are stored in a specific Database, every Bot will have its own space.

When uploading a new resource, the embedding status will be Pending, if the upload is succesfull the status switch to Indexed, if something goes wrong ( for example an unsupported file ) it become Not Indexed.

When using the resources feature, be aware that the number of tokens utilized will increase proportionally to the size of the document. The number of tokens used by your model will be directly reflected by the general usage and the actual cost of each prompt.

Usually, one token equals one word