Skip to knowledge simulation

SpiritBox knowledge interactive simulation

Interactive simulation: files, passages, counts, and status values are examples. No file is uploaded or indexed.

Preview limits
It shows the intended document and semantic-recall flow. It does not prove format support, extraction quality, storage behavior, or a live retrieval result.

Knowledge Base

Upload Documents

Example formats

This example illustrates extracting and indexing document text for semantic recall, then reviewing passages before adding selected context to a chat.

Drag & drop documents here
PDF DOCX TXT MD CSV XLSX JSON HTML RTF

Example Indexed Documents

Demo data
DocumentTypeStatus
Q4_Financial_Report.pdf PDF Indexed
Product_Requirements.docx DOCX Indexed
Customer_Feedback_2026.csv CSV Indexed
API_Documentation.md Markdown Indexed
Budget_Projections.xlsx Excel Indexed
Meeting_Notes_Feb.txt Text Indexed

How Knowledge Base Works

1. Upload Drop in your files
2. Chunk Smart text splitting
3. Embed pgvector indexing
4. Review Select useful passages

This simulation depicts document extraction, pgvector embedding, and PostgreSQL storage on user-controlled hardware. Use >>recall <topic> to retrieve relevant chunks and choose which ones enter the model context. Core knowledge storage is designed to remain local; optional model, search, messaging, update, and API connections may send selected data to external services when you enable them.

Semantic Recall

In chat, request matches with:

>>recall project schedule risks

SpiritBox returns matching chunks for review. You decide which passages to inject into the conversation.

Local Storage Boundary

Original filesWorkspace/Knowledge/documents
Search indexPostgreSQL + pgvector
Context injectionUser selected
Current limit: Scanned documents may need OCR before their text can be indexed. Check the current edition documentation for supported extraction paths.

Best Practices

  • Use clean, well-formatted documents for best results
  • PDF tables are extracted with pdfplumber for accuracy
  • CSV/Excel files are chunked row-by-row with headers
  • Markdown preserves code blocks and structure
  • Review how large documents are split before relying on retrieved passages
  • Re-upload a file to update its index
Exit demo