---
name: embedding_store_usearch
kind: function
lang: py
domain: infra
version: "1.0.0"
purity: impure
signature: "def embedding_store_usearch(path: str, ids: list, embeddings: list, dim: int = 384) -> int"
description: "Crea indice USearch con embeddings y lo persiste a archivo. Busqueda sub-milisegundo."
tags: [embedding, usearch, vector, store, ann, python]
uses_functions: []
uses_types: []
returns: []
returns_optional: false
error_type: "error_go_core"
imports: [usearch, numpy]
tested: false
tests: []
test_file_path: ""
file_path: "python/functions/embedding/usearch_store.py"
---

## Ejemplo

```python
model = embedding_load_model(".local/models/e5-small")
docs = ["La IA transforma la industria", "Python es versatil"]
embs = embedding_encode(model, docs, mode="document")

n = embedding_store_usearch("docs.usearch", [0, 1], embs)
# n = 2
```

## Notas

USearch usa HNSW (approximate nearest neighbors). Para 50k vectores dim=384:
~80 MB en disco, busqueda ~0.13ms/query (150x mas rapido que sqlite-vec).
El tradeoff es que no soporta metadata nativa — usar junto con SQLite para metadata.
Sobreescribe el archivo si ya existe.