~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
"""Shared SQLAlchemy connection pool configuration constants."""
# Per-user encrypted-database pool sizing. See ADR-0004 for why 20/40:
# inject_current_user() opens a session on essentially every request, so a
# small pool is exhausted by ordinary page loads.
#
# These are exported rather than inlined because the web layer needs to
# reason about the same numbers: sync routes retain one connection per AnyIO
# worker thread (see the threadpool sizing check in web/fastapi_app.py), so
# a worker pool larger than POOL_SIZE + MAX_OVERFLOW is a misconfiguration.
POOL_SIZE = 20
MAX_OVERFLOW = 40
# Validate connections before checkout (detects stale/broken connections)
POOL_PRE_PING = True
# Recycle connections after 1 hour to release stale file handles.
# SQLite reopens are cheap (no network roundtrip), so a shorter
# interval reduces the window for WAL handle accumulation.
# See ADR-0004.
POOL_RECYCLE_SECONDS = 3600