Saltar al contenido principal

RAG Files

RAG (Retrieval-Augmented Generation) files are documents you upload so the AI assistant can reference them when answering questions. This is ideal for existing documentation like PDFs, SOPs, and manuals.

How RAG files work​

  1. You upload a file in the Backoffice
  2. Appilot processes the file: extracts text, splits into chunks, generates embeddings
  3. When a user asks a question, relevant chunks are retrieved
  4. The AI uses these chunks alongside knowledge content to generate answers

Uploading files​

  1. Go to RAG Files in the Backoffice
  2. Click Upload File
  3. Select the file (PDF, TXT)
  4. Associate it with a domain and/or URL pattern
  5. Upload — processing begins automatically

File association​

Files can be scoped to:

  • Specific domain + URL pattern: Only used for questions on matching pages
  • Specific domain: Used for any question on that domain
  • Organization-wide: Used as general context for any question

Best practices​

Choose the right files​

Good candidates for RAG files:

  • Standard operating procedures (SOPs)
  • Product manuals and user guides
  • Policy documents
  • FAQ documents
  • Training materials

Keep files focused​

  • Smaller, focused documents produce better results than large monolithic files
  • Split a 200-page manual into topic-specific sections
  • Each file should cover a coherent topic

Keep files current​

  • Remove or replace outdated files — they cause incorrect answers
  • Set a review schedule (e.g., quarterly)
  • Use the file's "last updated" date as a freshness indicator

Scope files precisely​

  • A product manual should be associated with the product's domain
  • A general company policy can be organization-wide
  • An SOP for a specific form should be scoped to that form's URL pattern

Processing status​

After upload, files go through processing stages:

StatusMeaning
UploadingFile is being transferred
ProcessingText extraction and chunking in progress
EmbeddingVector embeddings are being generated
ReadyFile is available for RAG queries
ErrorProcessing failed — check file format and try again