Overview

The knowledge base indexes your files and data sources for retrieval‑augmented generation. Agents fetch relevant chunks, ground their answers, and include citations.

Sources

• Upload files (PDF, DOCX, TXT, CSV, PPTX, MD).
• Connect cloud drives (Google Drive) and select folders.
• Sync from URLs or sitemaps with the web crawler.

Syncing and versioning

• Automatic background sync keeps indexes fresh.
• Delta updates minimize re‑processing large sources.
• Manual reindex available per source and per file.

Chunking and embeddings

• Adaptive chunking based on structure (headings, lists, tables).
• Embeddings stored with metadata: file, page, section, tags.
• Configurable max tokens and overlap for your domain.

Retrieval and citations

• Hybrid search (semantic + keyword) improves recall.
• Per‑agent top‑k and score thresholds control grounding.
• Answers include clickable citations to original passages.

Permissions

• Project‑scoped access: agents only see sources added to the project.
• Optional per‑source visibility and redaction rules.

Best practices

• Prefer well‑structured documents; add headings for context.
• Tag sources by topic; agents can filter by tags.
• Keep large tables in CSV for accurate parsing.

Troubleshooting

• If citations look wrong, reduce chunk size or increase overlap.
• Re‑sync after major edits; stale chunks can surface.
• Check the processing log for extraction errors per file.