~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
/**
* Runtime contracts for the menu-to-settings bridge loaded by base.html.
*
* This module writes directly to the FastAPI single-setting route rather
* than going through services/api.js, so exercising the real change handler
* is the only way to bind its method, CSRF header, body, and response flow.
*/
const flushPromises = async (turns = 16) => {
for (let turn = 0; turn < turns; turn += 1) {
await Promise.resolve();
}
};
async function loadSettingsSync() {
const addEventListener = vi.spyOn(document, 'addEventListener');
await import('@js/components/settings_sync.js');
const registration = addEventListener.mock.calls.find(
([eventName]) => eventName === 'DOMContentLoaded',
);
addEventListener.mockRestore();
const handler = registration[1];
document.removeEventListener('DOMContentLoaded', handler);
handler();
}
function successfulResponse(data = { status: 'success' }) {
return new Response(JSON.stringify(data), { status: 200 });
}
beforeEach(() => {
vi.resetModules();
document.body.innerHTML = `
<input id="model_hidden" value="qwen3">
<select id="model_provider">
<option value="ollama" selected>Ollama</option>
</select>
<input id="search_engine_hidden" value="searxng">
<input id="iterations" value="3">
<input id="questions_per_iteration" value="4">
<select id="strategy">
<option value="source-based" selected>Source based</option>
</select>
<input id="ollama_url" value="http://ollama:11434">
<input id="custom_endpoint" value="https://llm.example/v1">
<select id="theme-select">
<option value="dark" selected>Dark</option>
</select>
`;
window.api = { getCsrfToken: vi.fn(() => 'csrf-current') };
window.displayWarnings = vi.fn();
window.refetchSettingsAndUpdateWarnings = vi.fn();
window.ui = { showMessage: vi.fn() };
vi.stubGlobal('requestIdleCallback', vi.fn(callback => callback()));
});
afterEach(() => {
vi.restoreAllMocks();
vi.unstubAllGlobals();
delete window.api;
delete window.displayWarnings;
delete window.refetchSettingsAndUpdateWarnings;
delete window.ui;
delete window.themeService;
document.body.replaceChildren();
});
it('persists a provider change through the FastAPI setting route and refreshes warnings', async () => {
const warnings = [{ type: 'provider', message: 'Local provider selected' }];
const fetchMock = vi.fn().mockResolvedValue(new Response(JSON.stringify({
status: 'success',
warnings,
}), { status: 200 }));
vi.stubGlobal('fetch', fetchMock);
await loadSettingsSync();
document.getElementById('model_provider').dispatchEvent(new Event('change'));
await flushPromises();
expect(fetchMock).toHaveBeenCalledOnce();
expect(fetchMock).toHaveBeenCalledWith('/settings/api/llm.provider', {
method: 'PUT',
headers: {
'Content-Type': 'application/json',
'X-CSRFToken': 'csrf-current',
},
body: JSON.stringify({ value: 'ollama' }),
});
expect(window.displayWarnings).toHaveBeenCalledWith(warnings);
expect(window.refetchSettingsAndUpdateWarnings).toHaveBeenCalledOnce();
expect(window.ui.showMessage).toHaveBeenCalledWith(
'provider updated successfully',
'success',
);
});
it.each([
{ detail: 'unsupported search engine' },
{
error: 'unsupported search engine',
detail: [{ loc: ['body', 'value'], msg: 'unsupported search engine', type: 'value_error' }],
},
])('surfaces a rejected FastAPI setting write without reporting success: %j', async body => {
const fetchMock = vi.fn().mockResolvedValue(new Response(
JSON.stringify(body),
{ status: 422, statusText: 'Unprocessable Entity' },
));
vi.stubGlobal('fetch', fetchMock);
await loadSettingsSync();
document.getElementById('search_engine_hidden').dispatchEvent(
new Event('change'),
);
await flushPromises();
expect(fetchMock).toHaveBeenCalledWith(
'/settings/api/search.tool',
expect.objectContaining({ method: 'PUT' }),
);
expect(window.displayWarnings).not.toHaveBeenCalled();
expect(window.refetchSettingsAndUpdateWarnings).not.toHaveBeenCalled();
expect(window.ui.showMessage).toHaveBeenCalledWith(
'Error updating search.tool: unsupported search engine',
'error',
);
});
it('maps every menu control to its individual FastAPI setting route', async () => {
const fetchMock = vi.fn().mockImplementation(() =>
Promise.resolve(successfulResponse()));
vi.stubGlobal('fetch', fetchMock);
await loadSettingsSync();
const controls = [
['model_hidden', 'llm.model', 'qwen3'],
['iterations', 'search.iterations', '3'],
['questions_per_iteration', 'search.questions_per_iteration', '4'],
['strategy', 'search.search_strategy', 'source-based'],
['ollama_url', 'llm.ollama.url', 'http://ollama:11434'],
['custom_endpoint', 'llm.openai_endpoint.url', 'https://llm.example/v1'],
];
for (const [id] of controls) {
document.getElementById(id).dispatchEvent(new Event('change'));
}
await flushPromises();
expect(fetchMock).toHaveBeenCalledTimes(controls.length);
controls.forEach(([id, key, value], index) => {
expect(fetchMock).toHaveBeenNthCalledWith(
index + 1,
`/settings/api/${key}`,
expect.objectContaining({
method: 'PUT',
body: JSON.stringify({ value }),
}),
);
expect(document.getElementById(id).value).toBe(value);
});
expect(window.refetchSettingsAndUpdateWarnings).toHaveBeenCalledTimes(3);
});
it('delegates theme changes to the theme service without a duplicate write', async () => {
const fetchMock = vi.fn();
vi.stubGlobal('fetch', fetchMock);
window.themeService = { setTheme: vi.fn() };
await loadSettingsSync();
document.getElementById('theme-select').dispatchEvent(new Event('change'));
expect(window.themeService.setTheme).toHaveBeenCalledWith('dark', true);
expect(fetchMock).not.toHaveBeenCalled();
});
it('falls back to the setting route when the theme service is unavailable', async () => {
const fetchMock = vi.fn().mockResolvedValue(successfulResponse());
vi.stubGlobal('fetch', fetchMock);
await loadSettingsSync();
document.getElementById('theme-select').dispatchEvent(new Event('change'));
await flushPromises();
expect(fetchMock).toHaveBeenCalledWith('/settings/api/app.theme', {
method: 'PUT',
headers: {
'Content-Type': 'application/json',
'X-CSRFToken': 'csrf-current',
},
body: JSON.stringify({ value: 'dark' }),
});
});
it('saves changed URL inputs once on blur and remembers the saved value', async () => {
const fetchMock = vi.fn().mockResolvedValue(successfulResponse());
vi.stubGlobal('fetch', fetchMock);
await loadSettingsSync();
const ollamaUrl = document.getElementById('ollama_url');
ollamaUrl.dispatchEvent(new Event('blur'));
ollamaUrl.dispatchEvent(new Event('blur'));
await flushPromises();
expect(fetchMock).toHaveBeenCalledOnce();
expect(fetchMock).toHaveBeenCalledWith(
'/settings/api/llm.ollama.url',
expect.objectContaining({
body: JSON.stringify({ value: 'http://ollama:11434' }),
}),
);
expect(ollamaUrl.getAttribute('data-last-saved'))
.toBe('http://ollama:11434');
ollamaUrl.value = '';
ollamaUrl.dispatchEvent(new Event('blur'));
expect(fetchMock).toHaveBeenCalledOnce();
});
it.each([
[
'Ollama URL',
'ollama_url',
'llm.ollama.url',
'http://ollama-new.example:11434',
],
[
'custom endpoint URL',
'custom_endpoint',
'llm.openai_endpoint.url',
'https://gateway-new.example/v1',
],
])('saves an edited %s once across change followed by blur', async (
_label,
inputId,
settingKey,
value,
) => {
const fetchMock = vi.fn().mockResolvedValue(successfulResponse());
vi.stubGlobal('fetch', fetchMock);
await loadSettingsSync();
const input = document.getElementById(inputId);
input.value = value;
// This is the event order emitted when an edited text input loses focus.
input.dispatchEvent(new Event('change'));
input.dispatchEvent(new Event('blur'));
await flushPromises();
expect(fetchMock).toHaveBeenCalledOnce();
expect(fetchMock).toHaveBeenCalledWith(`/settings/api/${settingKey}`, {
method: 'PUT',
headers: {
'Content-Type': 'application/json',
'X-CSRFToken': 'csrf-current',
},
body: JSON.stringify({ value }),
});
expect(input.getAttribute('data-last-saved')).toBe(value);
});
it('releases URL deduplication after a rejected save so blur can retry', async () => {
const fetchMock = vi.fn()
.mockResolvedValueOnce(new Response(JSON.stringify({
detail: 'endpoint is temporarily unavailable',
}), { status: 503 }))
.mockResolvedValueOnce(successfulResponse());
vi.stubGlobal('fetch', fetchMock);
await loadSettingsSync();
const input = document.getElementById('custom_endpoint');
input.value = 'https://retry.example/v1';
input.dispatchEvent(new Event('change'));
input.dispatchEvent(new Event('blur'));
await flushPromises();
expect(fetchMock).toHaveBeenCalledOnce();
expect(input.hasAttribute('data-last-saved')).toBe(false);
expect(window.ui.showMessage).toHaveBeenCalledWith(
'Error updating llm.openai_endpoint.url: ' +
'endpoint is temporarily unavailable',
'error',
);
input.dispatchEvent(new Event('blur'));
await flushPromises();
expect(fetchMock).toHaveBeenCalledTimes(2);
expect(input.getAttribute('data-last-saved'))
.toBe('https://retry.example/v1');
});
it('serializes same-key writes while unrelated settings save concurrently', async () => {
let resolveFirstModelSave;
const firstModelSave = new Promise(resolve => {
resolveFirstModelSave = resolve;
});
const fetchMock = vi.fn((url, options) => {
const { value } = JSON.parse(options.body);
if (url === '/settings/api/llm.model' && value === 'qwen3') {
return firstModelSave;
}
return Promise.resolve(successfulResponse());
});
vi.stubGlobal('fetch', fetchMock);
await loadSettingsSync();
const model = document.getElementById('model_hidden');
model.dispatchEvent(new Event('change'));
model.value = 'llama3.3';
model.dispatchEvent(new Event('change'));
document.getElementById('model_provider').dispatchEvent(
new Event('change'),
);
await flushPromises();
// The newer model value waits for the older model PUT to settle, but the
// independent provider key is not held behind that queue.
expect(fetchMock).toHaveBeenCalledTimes(2);
expect(fetchMock).toHaveBeenNthCalledWith(
1,
'/settings/api/llm.model',
expect.objectContaining({
body: JSON.stringify({ value: 'qwen3' }),
}),
);
expect(fetchMock).toHaveBeenNthCalledWith(
2,
'/settings/api/llm.provider',
expect.objectContaining({
body: JSON.stringify({ value: 'ollama' }),
}),
);
resolveFirstModelSave(successfulResponse());
await flushPromises(16);
expect(fetchMock).toHaveBeenCalledTimes(3);
expect(fetchMock).toHaveBeenNthCalledWith(
3,
'/settings/api/llm.model',
expect.objectContaining({
body: JSON.stringify({ value: 'llama3.3' }),
}),
);
});
it('uses the animation-frame bootstrap when idle callbacks are unavailable', async () => {
vi.stubGlobal('requestIdleCallback', undefined);
const animationFrame = vi.fn(callback => callback());
vi.stubGlobal('requestAnimationFrame', animationFrame);
vi.stubGlobal('fetch', vi.fn().mockResolvedValue(successfulResponse()));
await loadSettingsSync();
expect(animationFrame).toHaveBeenCalledOnce();
document.getElementById('model_hidden').dispatchEvent(new Event('change'));
await flushPromises();
expect(fetch).toHaveBeenCalledWith(
'/settings/api/llm.model',
expect.objectContaining({ method: 'PUT' }),
);
});