OpenParallax

Appearance

Planned — Not Yet Implemented

This backend is on the roadmap but not yet implemented. The interface is designed for future additions. Currently only SQLite is available.

ChromaDB ​

Simple, developer-friendly vector database popular in the AI/ML ecosystem.

Overview ​

ChromaDB is an open-source embedding database designed for simplicity:

  • Simple API -- minimal configuration, intuitive interface
  • HNSW index -- fast approximate nearest neighbor search
  • Metadata filtering -- filter by key-value metadata during search
  • Persistent storage -- SQLite-backed persistence
  • In-memory mode -- for testing and prototyping
  • Popular in AI/ML -- widely used with LangChain, LlamaIndex, and other AI frameworks

When to Use ​

  • AI/ML prototyping -- simple API gets you started fast
  • LangChain/LlamaIndex integration -- ChromaDB is a first-class citizen in these frameworks
  • Small to medium workloads -- up to ~1M vectors
  • Development and testing -- in-memory mode for unit tests
  • Python-heavy teams -- ChromaDB's ecosystem is Python-native

Setup ​

Self-Hosted ​

bash
# Docker
docker run -d --name chroma \
  -p 8000:8000 \
  -v chroma_data:/chroma/chroma \
  chromadb/chroma

# Or install the Python package for local mode
pip install chromadb

Create the Store ​

go
import "github.com/openparallax/openparallax/memory/chroma"

store, err := chroma.NewStore(chroma.Options{
    URL:            "http://localhost:8000",
    CollectionName: "memories",
})
if err != nil {
    log.Fatal(err)
}
defer store.Close()
python
# Python
from openparallax_memory import ChromaStore

store = ChromaStore(
    url="http://localhost:8000",
    collection_name="memories",
)
typescript
// Node.js
import { ChromaStore } from '@openparallax/memory'

const store = new ChromaStore({
  url: 'http://localhost:8000',
  collectionName: 'memories',
})

Configuration ​

go
store, _ := chroma.NewStore(chroma.Options{
    URL:            "http://localhost:8000",
    CollectionName: "memories",
    DistanceFunction: "cosine",  // "cosine", "l2", or "ip" (inner product)
    Tenant:         "default",   // multi-tenancy support
    Database:       "default",   // database isolation
})

Metadata Filtering ​

ChromaDB supports metadata filtering during search:

go
// Store with metadata
store.Upsert(ctx, "doc-1", embedding, map[string]string{
    "source": "meeting-notes",
    "date":   "2026-03-15",
})

// Search with filter
results, _ := store.Search(ctx, queryVector, 10,
    chroma.WithWhere(map[string]any{
        "source": "meeting-notes",
    }),
)

// Complex filters
results, _ := store.Search(ctx, queryVector, 10,
    chroma.WithWhere(map[string]any{
        "$and": []map[string]any{
            {"source": "meeting-notes"},
            {"date": map[string]any{"$gte": "2026-01-01"}},
        },
    }),
)

Supported operators:

  • $eq, $ne -- equality
  • $gt, $gte, $lt, $lte -- comparison
  • $in, $nin -- set membership
  • $and, $or -- boolean combinations

Full-Text Search ​

ChromaDB supports basic document search through its where_document filter:

go
results, _ := store.Search(ctx, queryVector, 10,
    chroma.WithWhereDocument(map[string]any{
        "$contains": "CORS middleware",
    }),
)

INFO

ChromaDB's document search is basic substring matching, not ranked full-text search. For BM25-style keyword search, Memory provides application-level hybrid search using a local FTS5 index.

Performance ​

VectorsQuery (top-10)Memory
10,000< 5ms~200 MB
100,000< 10ms~1.5 GB
500,000< 30ms~7 GB
1,000,000< 50ms~14 GB

Limitations ​

  • No native hybrid search -- vector search only; keyword search via substring matching
  • No managed cloud -- self-hosted only (as of 2026)
  • No ACID transactions -- eventual consistency
  • Scaling ceiling -- practical limit around 1M vectors per collection
  • Python-centric ecosystem -- Go and Node.js support is via REST API

Next Steps ​