Skip to main content
A Qdrant collection becomes nested folders, one level per group-by payload field. Each point is a .json payload file, plus a .txt text file and an optional blob.

Setup

A self-hosted Qdrant needs no key: set host and port (default localhost:6333) instead of url. Qdrant Cloud needs the cluster URL and an API key:
  1. In the Qdrant Cloud console, open the cluster and go to its API Keys tab.
  2. Click Create, name the key and pick an expiration (90 days by default).
  3. Pick read-only access, for the whole cluster or for the collections to mount.
  4. Copy the key, which shows only once, and the cluster URL.
A folder listing creates a keyword payload index for a group_by field only when Qdrant reports one is required. A read-only key cannot, so index those fields first or use a key that manages the collection.

Connect

workspace.yaml

Layout

  • Without collection, every collection is a top-level folder. Files are named by point id, or <name>__<id> with name_field.
  • Fields take dotted payload paths, such as group_by: [metadata.source], and basename_fields keeps only the last path segment of a value.
  • The search CLI ranks points by meaning. It needs an embed hook (Python and TypeScript) or cloud_inference: true, where the cluster embeds the query with embedding_model; with neither it refuses the query. The stored vectors must come from the same model.

Sample runs

The mount is read-only. One folder listing scrolls at most max_rows (1,000) points, and search returns search_limit (10) results by default.

SDK guides

Python

Every config field, the VFS.md and write support.

TypeScript

Every config field, the VFS.md and write support.