~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
Python academiaanthropicarxivbravedeep-researchencryptionhome-automationhomeserverlocallocal-deep-researchlocal-llmmistralollamaopenaipubmedresearchresearch-toolretrieval-augmented-generationsearxngself-hosted
"""Regression: VectorIndex must LOAD an existing on-disk index without error.
Guards a real bug where __init__ passed `path` both positionally and via the
kwargs dict to the store's load(), raising
`TypeError: got multiple values for argument 'path'` — which broke every search
and every 2nd index/delete on an existing collection.
"""
import threading
import numpy as np
from local_deep_research.vector_stores import get_vector_store_class
from local_deep_research.vector_stores.facade import VectorIndex
def _noop(_p):
return None
def _ok(_p):
return (True, None)
def test_vector_index_loads_existing_index(tmp_path):
path = tmp_path / "coll.faiss"
lock = threading.Lock()
dim = 8
# Pre-create a real on-disk IndexIDMap2.
store = get_vector_store_class().create(
dimension=dim,
index_type="flat",
metric="cosine",
normalize=True,
path=path,
lock=lock,
integrity_record=_noop,
integrity_verify=_ok,
)
store.apply(
add_ids=[11, 22], add_vectors=np.random.rand(2, dim).astype("float32")
)
assert path.exists()
# Construct VectorIndex over the EXISTING file — must load, not raise.
vi = VectorIndex(
username="u",
db_password="pw",
embeddings=None,
embedding_model="m",
embedding_model_type=None,
collection_name="c",
dimension=dim,
path=path,
lock=lock,
integrity_record=_noop,
integrity_verify=_ok,
index_type="flat",
metric="cosine",
normalize=True,
)
assert vi._store.count() == 2
assert set(vi._store.live_ids()) == {11, 22}