~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
"""Long API v1 research must leave the default auth/DB executor available."""
import asyncio
import threading
from concurrent.futures import ThreadPoolExecutor
from contextvars import ContextVar
from unittest.mock import patch
import pytest
from local_deep_research.config.thread_settings import (
get_settings_context,
set_settings_context,
)
from local_deep_research.web.dependencies import threadpool
@pytest.mark.parametrize(
("path", "target", "body", "result"),
[
(
"/api/v1/quick_summary",
"local_deep_research.api.research_functions.quick_summary",
{"query": "q"},
{"summary": "ok", "findings": []},
),
(
"/api/v1/generate_report",
"local_deep_research.api.research_functions.generate_report",
{"query": "q"},
{"content": "ok", "metadata": {}},
),
(
"/api/v1/analyze_documents",
"local_deep_research.web.routers.api_v1.analyze_documents",
{"query": "q", "collection_name": "Library"},
{"summary": "ok"},
),
],
)
def test_research_routes_use_the_dedicated_workers(
authenticated_client, path, target, body, result
):
observed_threads = []
def fake_research(*args, **kwargs):
observed_threads.append(threading.current_thread().name)
return result
with patch(target, side_effect=fake_research):
response = authenticated_client.post(path, json=body)
assert response.status_code == 200, response.text[:300]
assert len(observed_threads) == 1
assert observed_threads[0].startswith("ldr-api-research"), observed_threads
def test_research_does_not_starve_a_single_default_db_worker(monkeypatch):
entered = threading.Event()
release = threading.Event()
research_pool = ThreadPoolExecutor(max_workers=1)
monkeypatch.setattr(
threadpool, "_get_api_research_executor", lambda: research_pool
)
def hold_research_worker():
entered.set()
assert release.wait(timeout=10), "research worker was not released"
async def scenario():
loop = asyncio.get_running_loop()
loop.set_default_executor(ThreadPoolExecutor(max_workers=1))
research = asyncio.create_task(
threadpool.run_research_sync(hold_research_worker)
)
try:
for _ in range(200):
if entered.is_set():
break
await asyncio.sleep(0.01)
assert entered.is_set(), "research did not start"
# This models ensure_user_database on an authenticated request.
assert (
await asyncio.wait_for(
threadpool.run_db_sync(lambda: "auth-ready"), timeout=3
)
== "auth-ready"
)
finally:
release.set()
await asyncio.wait_for(research, timeout=10)
try:
asyncio.run(scenario())
finally:
release.set()
research_pool.shutdown(wait=True)
def test_research_dispatch_copies_request_context_and_clears_worker_state(
monkeypatch,
):
research_pool = ThreadPoolExecutor(max_workers=1)
monkeypatch.setattr(
threadpool, "_get_api_research_executor", lambda: research_pool
)
owner = ContextVar("api_research_owner", default=None)
async def scenario():
token = owner.set("alice")
try:
def first_call():
set_settings_context({"secret": "alice"})
return threading.get_ident(), owner.get()
first_thread, seen_owner = await threadpool.run_research_sync(
first_call
)
second_thread, settings = await threadpool.run_research_sync(
lambda: (threading.get_ident(), get_settings_context())
)
return first_thread, second_thread, seen_owner, settings
finally:
owner.reset(token)
try:
first_thread, second_thread, seen_owner, settings = asyncio.run(
scenario()
)
finally:
research_pool.shutdown(wait=True)
assert first_thread == second_thread
assert seen_owner == "alice"
assert settings is None