Chroma - Mirage

ChromaResource exposes a ChromaDB collection as a read-only filesystem: the directory tree comes from the collection’s __path_tree__ document, pages are chunk sets joined in chunk_index order, and vector retrieval is the chroma-query command. The TypeScript backend mirrors the Python one and returns identical results.

Node

The chromadb client is an optional peer dependency; install it alongside Mirage:

pnpm add @struktoai/mirage-node chromadb

import { ChromaResource, MountMode, Workspace } from '@struktoai/mirage-node'

const knowledge = new ChromaResource({
  config: {
    host: 'localhost',
    port: 8000,
    collectionName: 'knowledge',
    // slugField: 'page_slug',      (default)
    // chunkIndexField: 'chunk_index', (default)
  },
})

const ws = new Workspace({ '/knowledge': knowledge }, { mode: MountMode.READ })

console.log(await (await ws.execute('tree /knowledge/')).stdoutStr())
console.log(await (await ws.execute('grep -r refund /knowledge/policies/')).stdoutStr())
console.log(
  await (await ws.execute('chroma-query "how am I throttled" /knowledge/')).stdoutStr(),
)

grep pushes $contains / $regex filters down to ChromaDB before fine filtering locally, so recursive searches avoid fetching the whole collection.

​Node

Node