Skip to main content

​
What it is for

The knowledge base gives the agent a retrieval layer for documents that should inform answers but should not be hard-coded into tool descriptions. Good examples:
  • product documentation
  • onboarding guides
  • help center articles
  • university handbooks
  • internal SOPs
  • policy documents

​
What the user experience looks like

  1. Upload one or more documents.
  2. Wait for processing to finish.
  3. Enable the knowledge base.
  4. Warpy starts using those documents in answers.
The toggle stays disabled until at least one document is ready.

​
Processing pipeline

Upload
  -> Parse document
  -> Chunk content
  -> Create embeddings
  -> Store in pgvector

User question
  -> search_knowledge_base
  -> retrieve relevant chunks
  -> grounded answer

​
What happens under the hood

Warpy parses the uploaded file, chunks it, creates embeddings, and retrieves the most relevant sections at answer time.

​
Supported file types

Warpy supports common document, spreadsheet, image, and structured text formats, including PDF, DOCX, PPTX, TXT, MD, CSV, XLSX, HTML, JSON, and common image types. Maximum file size: 50 MB per file

​
Best practices

Use clean, current sources instead of stale exports or screenshots of documents.

Upload the docs the end user should actually rely on.

Wait until at least one document is ready, then test a few real questions.

Use the knowledge base to answer questions and tools to perform actions.

​
Operational details

  • The page auto-polls while documents are still processing.
  • Removing a document stops the agent from using it.
  • Ready documents can be previewed in the app.
  • Uploads may be blocked on plans with no remaining actions.

​
When to use it

Use the knowledge base when the agent should answer:
  • “How do I do this?”
  • “What does this policy mean?”
  • “What are the prerequisites?”
  • “What does the handbook say about this?”
Use tools when the user wants Warpy to actually perform the action.