Your search.
Your
infrastructure.

A Go vector database for integrating embeddings, running similarity search and keeping control of data and operations.

Community is a release candidate. Validate integration and data recovery before promoting an environment.

One query, over HTTPPOST

/vectors/search

{
  "values": [0.12, 0.84, 0.31],
  "k": 10
}

EmbeddingHNSWNeighbors

Index: Hierarchical HNSW
Alternative: Exact search
Persistence: Snapshot + WAL
Observability: Prometheus /metrics

Request example. Values must match the dimensions of indexed vectors.

The engine

What you need to integrate.
The control to operate.

Vector search

Layered HNSW for approximate search and an exact mode. Configurable construction and search parameters for your dataset.

Persistent data

Snapshot and WAL, with disk-backed payload storage options. Validate durability and recovery in your environment.

Direct integration

Vector creation and search, batch operations and pagination. Go, Python, JavaScript and Rust clients in the SDK repository.

Observable operations

Health, readiness and Prometheus metrics. Configurable API keys and TLS as part of your deployment controls.

Editions

Choose how to operate.

From self-hosted deployment to commercial options. Explore the availability and scope of each edition.

Community

Release candidate

MIT-licensed open source. Run on your own infrastructure. Current release: v0.3.0-rc.1.

Download release ↗

Business

By arrangement

Commercial distribution for specific operational and integration needs. Scope, support and terms agreed together.

Discuss Business ↗

Cloud

In preparation

Cluster management console and lifecycle workflows in development. The managed service is not yet open for public signup.

Ask about availability ↗

Quickstart / Community

From the repository
to your first query.

Build from source with Go 1.26 or newer, or download a release binary. Configure authentication and TLS before exposing the service.

Open full guide ↗

# Get and start the server
git clone https://github.com/brma-tech/lumenvec-community.git
cd lumenvec-community
go run ./cmd/server

# In another terminal
curl http://localhost:19190/health

HTTP :19190 · gRPC :19191 · One transport per process

LumenRAG / v0.1.7

From documents
to answers with sources.

A Studio for ingestion, retrieval, chat and evaluation on LumenVec. The distribution includes the engine and interface; you configure the AI provider.

Explore LumenRAG ↗

Install with Python 3.11+

python -m pip install lumenrag==0.1.7
lumenrag start

Wheels for Windows, Linux and macOS. Configure AI provider credentials to generate embeddings and answers.

PyPI ↗

Before you start

Practical questions.

Does LumenVec generate embeddings?

No. Your application or LumenRAG generates embeddings using a configured model and sends them to the database. LumenVec stores and searches those vectors.

Is HNSW a knowledge graph?

No. HNSW connects nearby vectors to support approximate search. Entity relationships and GraphRAG retrieval are a separate application layer.

Can I sign up for Cloud?

The managed service is still in preparation. Contact BRMA to discuss your requirements and availability. Community source and binaries are public.

Are there guaranteed performance figures?

No. Volume, dimensions, hardware, concurrency and retrieval quality affect results. Evaluate QPS, p95, recall, memory and construction cost against your workload.

The next conversation

What problem do
you need to solve?

Tell us the context, the constraints and what needs to change. We start there.

contact@brma.com.br

Your search.Yourinfrastructure.

What you need to integrate.The control to operate.