~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
#!/usr/bin/env python3
"""Prepare the library UI shard's real MiniLM model before offline tests.
Only this setup process requests public model acquisition. The seed and server
reuse the shared data-volume cache without relaxing their egress policy.
"""
import math
import os
def prepare_embeddings():
if not os.environ.get("LDR_DATA_DIR"):
raise ValueError(
"LDR_DATA_DIR must name the disposable CI data directory"
)
from local_deep_research.embeddings.embeddings_config import get_embeddings
from local_deep_research.utilities.resource_utils import safe_close
embeddings = get_embeddings(
provider="sentence_transformers",
model="all-MiniLM-L6-v2",
device="cpu",
settings_snapshot={
"policy.egress_scope": "public_only",
"search.tool": "arxiv",
"embeddings.require_local": False,
},
)
try:
# Load the actual weights/tokenizer, rather than accepting cache
# metadata alone as proof that an interrupted download completed.
vector = embeddings.embed_query("Synthetic library CI readiness check")
if len(vector) != 384 or not all(math.isfinite(x) for x in vector):
raise ValueError("The library fixture requires finite 384d vectors")
finally:
safe_close(embeddings, "library fixture embeddings")
print("Library MiniLM embedding cache is ready")
if __name__ == "__main__":
prepare_embeddings()