Lightweight vector database for learning vector search from first principles.

• Learn how vectors are stored, scored, and ranked without hidden magic.
• Small, readable codebase designed for tinkering.
• Easy to extend into ANN indexes later.

• Store high-dimensional embedding vectors.
• Brute-force similarity search (cosine, dot product).
• FastAPI HTTP API.
• File-based persistence (data/vectors.json).

uv sync --dev
uv run uvicorn app.main:app --reload

POST /vectors
{
  "id": "doc-1",
  "vector": [0.12, -0.04, 0.33, 0.91, -0.22, 0.05, -0.6, 0.78],
  "metadata": { "source": "openai" }
}

POST /search
{
  "query_vector": [0.12, -0.04, 0.33, 0.91, -0.22, 0.05, -0.6, 0.78],
  "top_k": 5,
  "metric": "cosine"
}