~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
"""
Test the browser's research creation endpoint specifically
"""
# allow: no-sut-import — this module exercises production code over HTTP
# rather than by importing it: every test drives real FastAPI routes
# (/api/start_research, /research/api/status/<id>) through the
# `authenticated_client` fixture, which builds the real app. There is no
# symbol to import and call directly; the routes ARE the subject. Verified
# not to be a shadow test in the sense the hook guards against: breaking
# /research/api/status/<id> in src/ makes test_research_status_endpoint
# fail, so production code really is under test here.
import json
class TestBrowserEndpoint:
"""Test browser-specific endpoints."""
def test_start_research_endpoint(self, authenticated_client):
"""Test the browser endpoint /api/start_research."""
research_data = {
"query": "Test from browser endpoint",
"mode": "quick",
"model_provider": "OLLAMA",
"model": "llama2",
"search_engine": "searxng",
"max_results": 10,
"time_period": "y",
"iterations": 1,
"questions_per_iteration": 3,
"strategy": "source-based",
"local_context": 2000,
"web_context": 2000,
"temperature": 0.7,
}
response = authenticated_client.post(
"/api/start_research",
json=research_data,
content_type="application/json",
)
assert response.status_code in [200, 202]
data = json.loads(response.data)
if response.status_code == 200:
assert data.get("status") in ["success", "processing"]
if "research_id" in data:
assert isinstance(data["research_id"], (str, int))
def test_research_status_endpoint(self, authenticated_client):
"""Test research status endpoint.
Was previously all-conditional: a 200 (or a research_id-less
200, or a 404 from the status lookup) skipped every assertion
and still passed. This "audit: ... issue resolved by prior PR"
marker was added by a comment-only bulk annotation (PR #4296,
explicitly "zero behavioral effect") — the underlying gap was
never actually fixed. Empirically /api/start_research returns
200 with a research_id, and the freshly created research_id
resolves via /research/api/status/<id> with 200, in this test
environment — so those are now asserted unconditionally.
"""
research_data = {
"query": "Test research for status check",
"mode": "quick",
"model_provider": "OLLAMA",
"model": "llama2",
"search_engine": "wikipedia",
"iterations": 1,
}
response = authenticated_client.post(
"/api/start_research",
json=research_data,
content_type="application/json",
)
assert response.status_code == 200, (
f"unexpected status starting research: {response.status_code} "
f"{response.data!r}"
)
data = json.loads(response.data)
research_id = data.get("research_id")
assert research_id, (
f"start_research response is missing research_id: {data!r}"
)
# Check status
status_response = authenticated_client.get(
f"/research/api/status/{research_id}"
)
assert status_response.status_code == 200, (
"status endpoint returned "
f"{status_response.status_code} for a research_id obtained "
"from start_research"
)
status_data = json.loads(status_response.data)
assert "status" in status_data
def test_endpoint_requires_authentication(self, client):
"""Test that endpoint requires authentication."""
research_data = {
"query": "Test without auth",
"mode": "quick",
}
response = client.post(
"/api/start_research",
json=research_data,
content_type="application/json",
)
# Should either redirect to login or return 401
assert response.status_code in [
302,
401,
403,
] # 403 = CSRF rejection (Wave 2 fail-closed)
if response.status_code == 302:
assert "/auth/login" in response.location