Vectra is a high-performance, API-key-based vector database designed for fast and meaningful semantic search. It enables users to generate embeddings, store vectors, and perform similarity searches using natural language queries without relying on traditional SQL-style filtering.
Vectra is built for developers who want semantic understanding, speed, and control over their vector data.
Vectra is a self-hosted vector database system that:
-
Uses HNSW (Hierarchical Navigable Small World) graphs for fast vector similarity search
-
Generates embeddings using a lightweight local embedding model
-
Allows users to:
- Register and generate an API key
- Create and manage their own vectors
- Perform semantic search via API endpoints
-
Returns either:
- The stored object itself, or
- The object ID, allowing integration with the user’s own database
The goal is to provide much faster and more semantically accurate search compared to traditional SQL LIKE or rule-based filtering queries.
Traditional queries struggle with meaning.
Example:
Query:
"Hourly wage not less than $13"
With classic SQL, this requires rigid filters and predefined conditions. With Vectra, the same query is converted into a vector and matched semantically against stored data, returning the correct results without explicit field-based filtering.
This approach is:
- More flexible
- More expressive
- Often significantly faster on large datasets
Vectra has been tested with large, separate datasets, consistently outperforming traditional querying approaches in both speed and relevance.
Vectra is intentionally designed as an API-key–based service, not as an in-process library.
The core idea behind this approach is simple:
All vector operations should happen outside the main application.
Currently, all major operations are handled through a small set of HTTP endpoints, allowing applications to interact with Vectra using simple HTTP requests.
This design enables:
- Centralized vector creation
- Centralized semantic search
- Reduced computational load on main applications
- Language-agnostic integration (any app that can send HTTP requests)
At its core, Vectra is built around three fundamental operations:
- Vector creation (embedding generation)
- Vector storage / indexing
- Semantic search
All interactions happen through API endpoints, authenticated via an API_KEY.
This means any application can:
- Send raw data to Vectra
- Let Vectra generate vectors
- Query semantic meaning later without embedding logic in the main app
The long-term goal is to deploy Vectra as a standalone vector service that can be consumed by:
- Single applications
- Multiple independent applications
- External clients
All without increasing the load or complexity of the main systems.
While the current implementation is still evolving, the architectural vision includes:
- Proper multi-tenant support
- User-specific vector isolation
- Optimized and configurable index management
- More advanced
API_KEY-based authorization strategies
In future iterations, this system could evolve into:
-
A more complex, optimized API-key-based platform
-
A microservices-oriented architecture
-
Or even distributed vector generation and search services, where:
- Embedding generation
- Indexing
- Searching are deployed independently and scaled separately
This would allow Vectra to handle significantly larger workloads and concurrent usage scenarios.
One of the motivations behind keeping Vectra simple at its core is to demonstrate how powerful vector databases can be even in minimal form.
To showcase this, a sector-specific, AI-integrated project built on top of Vectra is currently under development.
This upcoming project aims to demonstrate:
- Practical, real-world semantic search use cases
- Deep AI integration powered by Vectra’s vector infrastructure
- How meaningful results can be achieved without complex rule-based querying
🚧 Under development A real, production-oriented project powered by Vectra is coming soon. It will be added here once ready.
Coming soon.
Vectra uses an API key–based authentication system.
Flow:
- User registers
- An API key is generated
- All vector-related operations are performed using this API key
- Each user logically owns their vectors and searches
⚠️ Note: User-specific index isolation and index storage strategies are still under active development.
- 🔐 API-key-based access
- 🧬 Local embedding generation (no external API dependency)
- ⚡ HNSW-powered similarity search
- 📦 Object or ID-based search results
- 🧠 Natural language semantic querying
- 🏎️ Optimized for speed and relevance
Vectra is not production-ready yet. Some important features are still missing or incomplete:
- ❌ Per-user isolated index files
- ❌ Proper index classification and storage strategy
- ❌ Production-grade deployment setup
- ❌ Public API access & rate limiting
- ❌ Persistence and lifecycle management for indexes
These are planned improvements.
Planned next steps include:
- User-specific index creation and isolation
- Configurable index storage paths
- Improved index metadata management
- Deployment-ready configuration
- API usage limits and monitoring
- Better documentation and usage examples
- Python
- HNSW for vector similarity search
- Local embedding model
- Custom API layer
- File-based index storage (for now)
Vectra is currently a research and development project focused on:
- Exploring fast semantic search
- Evaluating vector-based querying over traditional databases
- Building a flexible, developer-friendly vector database API
It is actively evolving and not yet open for public production use.
Contributions, discussions, and ideas are welcome. This project is still experimental, so feedback is highly valuable.
License information will be added.
Vectra Fast, semantic, and developer-controlled vector search.