Anvia

Appearance

@anvia/chroma ​

@anvia/chroma stores Anvia embedded documents in ChromaDB and exposes them through the SDK's vector-search interface.

Install ​

sh
pnpm add @anvia/chroma @anvia/core @anvia/openai chromadb

The package is ESM-only, includes chromadb, and peers with @anvia/core >=0.7.1 <1.0.0.

Store and search documents ​

ts
import { embedDocuments } from '@anvia/core/embeddings'
import { ChromaVectorStore } from '@anvia/chroma'
import { OpenAIClient } from '@anvia/openai'

const openai = new OpenAIClient({
  apiKey: process.env.OPENAI_API_KEY,
})
const embeddings = openai.embeddingModel('text-embedding-3-small')
const sourceDocuments = [
  {
    id: 'password-reset',
    text: 'Password reset links expire after 30 minutes.',
    category: 'account',
  },
]

const documents = await embedDocuments(embeddings, sourceDocuments, {
  id: (document) => document.id,
  content: (document) => document.text,
  metadata: (document) => ({ category: document.category }),
})

const store = await ChromaVectorStore.connect({
  collectionName: 'support_docs',
})

await store.upsertDocuments(documents)

const results = await store.index(embeddings).search({
  query: 'How do I reset a password?',
  topK: 5,
})

The index also implements searchIds() and asTool(). See Vector stores and Search tools.

Collection ownership ​

With the default createIfMissing: true, connect() gets or creates the collection. It supplies embeddingFunction: null because Anvia writes precomputed embeddings, and defaults collection metadata to cosine space.

For production, provision the collection with your infrastructure workflow and use:

ts
const store = await ChromaVectorStore.connect({
  client,
  collectionName: 'support_docs',
  createIfMissing: false,
})

Pass metadata or configuration when Studio should create the collection with explicit Chroma settings. Keep those settings compatible with the embedding model used to produce stored vectors.

Production patterns ​

  • Inject an authenticated, application-managed Chroma client instead of relying on the no-argument default client.
  • Keep collection creation in deployment infrastructure and fail startup when it is missing.
  • Preserve stable document IDs so repeated ingestion updates the same vector records.
  • Reserve capacity and retention around the number of embeddings, not only source documents.
  • Apply metadata filters for tenant or corpus boundaries; filters are not a substitute for authorization.

Reference ​

Built for Anvia.