OrcAI
Using OrcAI

Content Library

Upload, manage, process, and share reusable source material.

The Content Library (Library > Content) is where OrcAI stores reusable source material used for retrieval, grounding, and citations.

What You Can Do

  • Add content items
  • View content item details
  • Edit metadata
  • Reprocess content
  • Manage access and groups
  • Download original files (if permitted)
  • Delete content (if permitted)

End-To-End Flow

  1. Upload material to OrcAI.
  2. Confirm metadata in the intake flow.
  3. Store source in S3-compatible object storage.
  4. Process content through background jobs.
  5. Index retrieval-ready chunks.
  6. Attach items to repository blocks.
  7. Use repository blocks in bots or directly in chats.

Add Content UX

The Add Content page uses a multi-step intake flow and emphasizes that metadata follows the content item wherever it is reused.

Placeholder text in the UI indicates intended source types include:

  • Documents/files
  • URLs
  • Text snippets

TODO

Confirm which of URL and snippet ingestion are fully active in production paths, not just planned by copy.

Processing Status And Reprocessing

Content cards expose processing status, and reprocess actions are only available when status is not pending/active and the user has edit capability.

Operationally:

  • Reprocess is useful after parser/config changes.
  • Reprocess can be triggered from list cards and detail pages.

TODO

Add a full status legend once all status transitions are documented centrally.

Permissions And Access

Content details and actions are capability-based:

  • download
  • edit
  • delete
  • manage_access

Access management opens a shared Access dialog for visibility, groups, and direct grants.

Relationship To Retrieval

Content items are not retrieved directly by default in bot workflows. They are typically connected through database blocks (repositories), then attached to bots or chats.

Platform Dependencies

At system level, content workflows rely on:

  • S3-compatible storage
  • Worker processing pipeline
  • Embedding generation
  • Qdrant indexing

The embedding model and vector dimensions must stay compatible with existing collections.

Embedding compatibility

Changing EMBEDDING_MODEL or EMBEDDING_DIMENSIONS after assets have already been indexed is not supported right now. Recreate the Qdrant collection and reprocess assets instead.

TODO

Document file size limits, per-type parsing limits, OCR behaviour for image-heavy PDFs in mixed-language files, and troubleshooting for failed ingestion or stuck processing states.

On this page