Library

Myeline has no manual file-upload feature. A user's library (and an organisation's library) is fed exclusively by two automatic channels:

• Cloud connectors — Google Drive, OneDrive, Dropbox, kDrive, Zotero, S3 / S3-compatible, WebDAV
• RSS / web scrapers

This is a design choice: keep the source of truth on drives the customer already controls, with no content duplication inside Myeline.

Scopes

/user/ma-bibliotheque aggregates your personal documents. /org/<slug>/workspace aggregates those shared inside an organisation. Each scope has its own ChromaDB collection:

Scope	ChromaDB collection	Visibility
Personal	user_<id>_personal	You only
Organisation	org_<id>_shared	Org members
Public platform	shared	Every user

Cloud connectors

Connect your drive from /user/cloud (or /org/<slug>/cloud on the organisation side):

• Google Drive, OneDrive, Dropbox, kDrive (sovereign-hybrid only, see Cloud connectors)
• S3 / S3-compatible (all editions)
• WebDAV — Nextcloud, ownCloud (all editions)
• Zotero (sovereign-hybrid only, indexes bibliographic metadata)

The check_cloud_sync cron (every 4 h, tier-dependent floor) detects new files and adds them to your collection. You can also trigger a manual sync (max once per hour).

Indexed formats

Format	Extensions	Notes
PDF	.pdf	text extraction via pdfplumber
Word	.docx	via python-docx
OpenDocument	.odt	via odfpy
Plain text	.txt, .md	UTF-8 expected
HTML	.html, .htm	sanitised via bleach

Files in other formats (XLSX, PPTX, ZIP, video, image…) are silently ignored by the indexer. Size limit: 50 MB per file (configurable via MAX_UPLOAD_SIZE_MB). OCR on scanned PDFs is disabled by default (enable via RAG_OCR_ENABLED=true in sovereign-hybrid only, Tesseract dependency).

RSS / web scrapers

/user/scrapers (Pro+) lets you watch RSS feeds or web pages — each fetched article is added to your personal library. An organisation member can "push" a scraper to the shared library (🏢 button on the scrapers page).

The check_user_scrapers cron runs 3 times per day at off-peak hours (03:00 / 07:00 / 23:00 UTC) to avoid saturating the server during business hours.

Indexing pipeline

New file detected (cloud sync) or article fetched (scraper)
  → MIME detection (magic bytes)
  → Text extraction
  → Chunking (~500 tokens, overlap 50)
  → bge-m3 embedding (local Ollama)
  → ChromaDB insert into the target collection
  → Document RAG-queryable

Indexing is async (RQ worker). On a 100-page PDF, count 30-90 seconds after sync.

Removal and re-indexing

• Unindex a document from the library: removes ChromaDB chunks. The file stays in the source drive.
• Re-index a full collection: /admin/library/reindex on the global admin side (costly, useful after embedding model upgrade).

Single-document chat

Clicking a document (personal or org) opens a chat scoped to that single document — useful to drill into a long PDF without polluting other sources. See RAG search § single-document.

Why no manual upload?

Three reasons:

1. Single source of truth — when a document evolves on the customer side (new PDF in Drive), Myeline picks it up automatically at the next sync. No drift between the uploaded version and the "official" version.
2. Compliance — the document stays in the customer's classification system (Drive, SharePoint, Nextcloud…) which already handles permissions, backups, retention.
3. Simplicity — a single ingestion path to maintain and monitor (cloud sync), rather than doubling up with an upload form that has its own bugs and limits.

If your use case still requires a manual upload (e.g. one-off documents unrelated to a drive), reach out to us — technically doable, to arbitrate per context.