Skip to main content
A ChromaDB collection becomes a tree of text files. Each file is its chunks joined in order.

Setup

Run a Chroma server (it listens on port 8000 by default) and store the collection in the shape the mount reads: one path tree document plus chunk documents that name their file.
Store one document with the ID __path_tree__. Its content is a JSON object, plain or gzip-compressed and base64-encoded, whose keys are file paths below the mount:
  • Forward slashes make folders.
  • No empty path, and no ., .. or empty segment.
  • Every path stays unique after leading and trailing slashes are trimmed.
  • A path is never both a file and a folder prefix: guides and guides/quickstart.md cannot both be files.
  • size, created_at and updated_at are optional and used when present.
Each chunk document names its file and its place in that file:
The keys default to page_slug and chunk_index. A collection with other names sets slug_field and chunk_index_field in the config.
Confirm the tree document and at least one chunk for a path:

Connect

workspace.yaml

Layout

  • Paths come from the __path_tree__ document; ssl: true reaches the server over HTTPS.
  • A file is its chunks sorted by chunk index and joined by line breaks, so no grep match spans two chunks.
  • The search CLI ranks chunks by meaning with Chroma’s vector query.

Sample runs

The mount is read-only. grep -r reads only the files Chroma finds the text in; Chroma matches case, so grep -ri reads every file.

SDK guides

Python

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

TypeScript

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