ovos-qdrant-embeddings-plugin

Qdrant-backed EmbeddingsDB vector store plugin for OpenVoiceOS.

Install

pip install ovos-qdrant-embeddings-plugin

What is EmbeddingsDB?

EmbeddingsDB is the abstract vector-store interface defined by ovos-plugin-manager (opm.embeddings). It provides a uniform API for storing, retrieving, and nearest-neighbour querying of embedding vectors regardless of the backend.

OVOS components that need semantic search (e.g. memory, skills, pipelines) discover the active backend via the OPM entry-point group opm.embeddings. This plugin registers itself as:

opm.embeddings = ovos-qdrant-embeddings-plugin

It pairs naturally with embedding producers such as ovos-gguf-plugin that generate the vectors you store here.

Quickstart: in-memory DB

import numpy as np
from ovos_qdrant_embeddings import QdrantEmbeddingsDB

# In-memory: no host, no path. Use it for development and CI.
db = QdrantEmbeddingsDB(config={"vector_size": 4})

# Store vectors
db.add_embeddings("apple",  np.array([1.0, 0.0, 0.0, 0.0]))
db.add_embeddings("banana", np.array([0.0, 1.0, 0.0, 0.0]))
db.add_embeddings("cherry", np.array([0.0, 0.0, 1.0, 0.0]))

# Nearest-neighbour query, returns [(key, score), ...]
results = db.query(np.array([0.9, 0.1, 0.0, 0.0]), top_k=2)
print(results)  # [('apple', 0.999...), ('banana', 0.099...)]

Configuration

Three client modes

In-memory (development or CI): neither host nor path set.

config = {"vector_size": 384}

Local persistent: data survives restarts.

config = {"path": "/var/lib/ovos/qdrant", "vector_size": 384}

Remote: connects to a running Qdrant server or Qdrant Cloud.

config = {
    "host": "my-qdrant.example.com",
    "port": 6333,
    "api_key": "my-secret-key",
    "vector_size": 384,
}

When to choose Qdrant over ChromaDB

• You need to run the vector store as a separate network service (microservice / homelab).
• Your collection grows to millions of vectors. Qdrant's HNSW index scales well at that size.
• You want Qdrant Cloud managed hosting.
• You need gRPC for high-throughput batch ingestion.

For a single-device OVOS installation with moderate data, the ChromaDB plugin may be simpler. Both expose the same EmbeddingsDB interface, so switching is a config change.

Further reading

• docs/configuration.md: all config keys, client modes, and the cosine normalization note.
• docs/usage.md: collections, CRUD, batch operations, metadata, and query.
• examples/quickstart.py: add and query vectors in memory.
• examples/collections.py: a multi-collection workflow.
• examples/remote_server.py: remote Qdrant setup.

Testing

Tests use an in-memory Qdrant client. No server is required.

pip install ovos-qdrant-embeddings-plugin[test]
pytest test/ -v

Credits

Originally developed by TigreGótico for OpenVoiceOS, sponsored by VisioLab. Modernized under the NGI0 Commons Fund / NLnet.

This work was sponsored by VisioLab, part of Royal Dutch Visio, is the test, education, and research center in the field of (innovative) assistive technology for blind and visually impaired people and professionals. We explore (new) technological developments such as Voice, VR and AI and make the knowledge and expertise we gain available to everyone.

This project was funded through the NGI0 Commons Fund, a fund established by NLnet with financial support from the European Commission's Next Generation Internet programme, under the aegis of DG Communications Networks, Content and Technology under grant agreement No 101135429.