Analytics preference

We only load Google Analytics after you opt in. You can change this choice at any time from Cookie settings.

Vectorstores Tab

The Vectorstores tab manages vector stores used by RAG nodes. Create stores, upload documents, and share them with your team.

The Vectorstores tab manages vector stores used by RAG nodes. Create stores, upload documents, and share them with your team.

Creating a Vector Store

  1. Click New Vector Store
  2. Enter a name and optional description
  3. Select a vector store credential – RAG: Qdrant + OpenAI (external Qdrant server), RAG: Psql + OpenAI (Heym's own Postgres database via pgvector), or RAG: Custom Embeddings (any OpenAI-compatible embedding endpoint, targeting either backend). The credential determines the store's backend.
  4. Optionally set a custom collection name (defaults to auto-generated)
  5. Save

A store's backend is fixed by its credential at creation time. In a RAG node, the Database dropdown filters stores to the matching backend.

Uploading Documents

  • Upload files – Drag and drop or pick several files. A store takes:
    • Documents – PDF, Markdown, text, CSV and JSON
    • Images – PNG, JPEG, TIFF, WebP and BMP, read with the same Tesseract OCR as the Converter node, language detected automatically
    • Audio – mp3, wav, ogg, opus (WhatsApp voice notes) and flac, transcribed with the Converter's Whisper model on your own instance; each chunk keeps where it starts and ends in the recording (start_seconds, end_seconds)
    • Any other plain-text file – a file such as notes or README is read as text when it is valid UTF-8
  • Scanned PDFs – pages without a text layer are read with OCR, up to 200 such pages per file
  • Background uploads – audio and PDFs with scanned pages take minutes, so Heym reads them in the background. They show at the top of the store's items as Transcribing or Reading scanned pages and turn into a normal source when their chunks are in. A failed one shows why and can be dismissed. With Override existing files, the old version stays searchable until the new one is ready
  • Duplicate handling – Choose to override or skip duplicate filenames
  • Progress – Upload progress is shown during ingestion

An image without readable text, an audio file without speech, and a binary file of another type are refused with a message.

Managing Content

  • View items – See source groups and document counts per store. Uploaded files group under their filename; documents a RAG node stored without a source of their own group under Added workflow: name, naming the workflow that wrote them
  • Delete sources – Remove specific files or source groups from a store, including everything one workflow added
  • Edit store – Change name, description, or credential
  • Search – POST /api/vector-stores/{id}/search with a query (and a limit from 1 to 20, default 5) returns the chunks a RAG node's search would return, best match first, each with its score and source. Heym Work's RAG screen uses it to test retrieval

Sharing

  • Share vector stores with other users by email
  • Shared stores appear with an indicator
  • Revoke sharing from the store card menu

Using in Workflows

In a RAG node, select the vector store by name. The node retrieves relevant chunks and augments the LLM context with them.