Vektor is a purely file-based, embedded Vector Database written in native PHP. It is designed for "Zero-RAM Overhead" operation, meaning it doesn't load the entire dataset into memory. Instead, it relies on strict binary file layouts and disk-based seeking to perform approximate nearest neighbor searches using the HNSW (Hierarchical Navigable Small World) algorithm.

Vektor supports optional Bearer Token authentication.

1. Create a .env file in the project root:

   cp .env.example .env

2. Set your secret token:

   VEKTOR_API_TOKEN=your-super-secret-token

• If VEKTOR_API_TOKEN is set, all endpoints (except /up) require the header: Authorization: Bearer <token>.
• If VEKTOR_API_TOKEN is not set, the API is open (public).

The system exposes a simple REST API.

Inserts a new vector into the database.

• Payload: {"id": "doc-1", "vector": [0.1, ...]}

Searches for the nearest neighbors of a given vector.

• Payload: {"vector": [0.1, ...], "k": 5}
• Response:

  {
      "results": [
          { "id": "doc-1", "distance": 0.98 },
          { "id": "doc-4", "distance": 0.85 }
      ]
  }

Deletes a vector from the database.

• Payload: {"id": "doc-1"}

Vacuum deleted vectors and rebalance the HNSW graph. This operation locks the database.

• Payload: None
• Response: {"status": "success", "message": "..."}

Returns DB statistics.

• Response:

  {
      "storage": { "vector_file_bytes": ... },
      "records": { "vectors_total": ... },
      "config": { "dimension": 1536, ... }
  }

Returns simple status for monitoring. Public endpoint.

• Response: {"status": "up"}

Requires PHP 8.2+ and Composer.

composer install --no-dev
chmod -R 777 data/

Point your web server's document root to the public/ directory, using the provided .htaccess (Apache) or nginx.conf.example (Nginx).

MIT