​

QiCapture

​

Overview

​

Responsibilities

• Clean and parse raw intake.
• Index and extract metadata from documents (via OCR or scripts).
• Chunk and embed clean text for semantic search.
• Form entity relationships and graphs from text.
• Synchronize indices and manage backups.

​

Flows

Intake raw file 
  -> profile & inspect 
  -> clean & normalize 
  -> produce staging import 
  -> Supabase import

​

Folder Structure & Table of Contents

​

Intake Pipeline Stages

• 10 Ingestion: Raw intake from various sources (Paperless, drive imports, manual uploads, email exports, web clips).
• 20 Extraction: OCR processing, text extraction, markdown exports, and transcripts.
• 30 Chunking: Processing clean text into chunks, manifest records, and source mapping.
• 40 Embeddings: Embedding models, Qdrant setup, collections, and vector quality tests.
• 50 Graphs: Entity extraction, relationship graphs, and entity maps.
• 60 Retrieval: Semantic search profiles, query logs, retrieval tests, and context packaging.
• 70 Memory: facts storage, session summaries, user preference maps.
• 80 Sync & Backups: Snapshot logs, restore points, and synchronization manifests.

​

Tools & System References

• NocoDB: Intake tracking database interface.
• Obsidian QiDocs: Vault indexing rules and guidelines for local notes.
• WikiJS: Ingest documentation library.
• Paperless-ngx: Document management setup and indexing rules.