~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
/**
* Settings Error Detection UI Test
*
* Tests settings page for error messages that appear when changing values.
* Monitors console errors and checks for error elements on the page after
* making changes to settings. This helps identify validation errors or
* save failures.
*
* What this tests:
* - Console error detection during setting changes
* - Network error monitoring (4xx/5xx responses)
* - DOM error element detection (.error, .alert-danger, etc.)
* - Setting input interaction (dropdowns, text fields)
*
* Prerequisites: Web server running on http://127.0.0.1:5000
*
* Usage: node tests/ui_tests/test_settings_errors.js
*/
const puppeteer = require('puppeteer');
const AuthHelper = require('./auth_helper');
const { getPuppeteerLaunchOptions } = require('./puppeteer_config');
const { expandAllSettingsSections } = require('./test_lib');
async function testSettingsChange() {
const browser = await puppeteer.launch(getPuppeteerLaunchOptions());
const page = await browser.newPage();
const baseUrl = 'http://127.0.0.1:5000';
const authHelper = new AuthHelper(page, baseUrl);
// Monitor console errors
page.on('console', msg => {
if (msg.type() === 'error') {
console.log('❌ BROWSER ERROR:', msg.text());
}
});
// Monitor network errors
page.on('response', response => {
if (response.status() >= 400) {
console.log('❌ NETWORK ERROR:', response.status(), response.url());
}
});
let failed = false;
try {
console.log('🔧 Testing settings change functionality...');
await authHelper.ensureAuthenticatedWithTimeout();
await page.goto('http://127.0.0.1:5000/settings/', {
waitUntil: 'domcontentloaded',
timeout: 30000
});
// Wait for page to load
await new Promise(resolve => setTimeout(resolve, 3000));
// Try to find and change a simple setting
console.log('🔍 Looking for a setting to change...');
// Find a dropdown or input field to change
const settingInput = await page.$('select[data-key], input[data-key]');
if (settingInput) {
console.log('✅ Found setting input, attempting to change...');
// Get the current value
const currentValue = await page.evaluate(el => el.value, settingInput);
console.log('Current value:', currentValue);
// Try to change it
await page.evaluate(el => {
if (el.tagName === 'SELECT') {
// For dropdown, select a different option
if (el.options.length > 1) {
el.selectedIndex = el.selectedIndex === 0 ? 1 : 0;
el.dispatchEvent(new Event('change', { bubbles: true }));
}
} else {
// For input, change the value
el.value += '_test';
el.dispatchEvent(new Event('change', { bubbles: true }));
}
}, settingInput);
console.log('⏳ Waiting for any errors after change...');
await new Promise(resolve => setTimeout(resolve, 2000));
} else {
console.log('❌ No setting inputs found');
// Both branches below used to only log — this file exited 0
// regardless of what it found, so it never actually detected
// anything. Not finding a single `[data-key]` setting input on
// the settings page is itself a real failure (the thing this
// test exists to exercise never ran), so it is wired to `failed`.
failed = true;
}
// Check for any error messages on the page.
//
// Only VISIBLE errors carrying text count. The selector below is
// deliberately broad — `[class*="error"]` matches anything with
// "error" anywhere in its class — and the templates ship several
// always-present, initially-empty containers that it hits:
// `ldr-error-message`, `ldr-error-text`, `ldr-field-error`,
// `ldr-input-help ldr-text-error`. `page.$$` returns those regardless
// of visibility, so failing on mere presence would fail every run
// whatever the page did — swapping a test that could never fail for
// one that could never pass, which is no better.
// ...and "visible" is only meaningful once the sections are open.
// Field-level errors are created on demand by settings.js's
// `markInvalidInput`, which already reveals the section it just
// wrote into, and a top-level `showAlert` banner writes into
// `#settings-alert` / `#filtered-settings-alert`, both outside every
// `.ldr-settings-section-body`. So neither real error source
// strictly depends on this expand to be visible. It is defensive:
// without it, an error container that does end up inside a
// collapsed body could still be hidden from the filter below by
// `offsetParent === null`, letting the suite pass no matter what the
// page did.
const expandedSections = await expandAllSettingsSections(page);
console.log(`🔓 Expanded ${expandedSections} collapsed settings section(s) before checking`);
const errorElements = await page.$$('.error, .alert-danger, .text-danger, [class*="error"]');
const shownErrors = [];
for (const el of errorElements) {
const info = await page.evaluate((node) => {
const style = window.getComputedStyle(node);
const hidden =
style.display === 'none' ||
style.visibility === 'hidden' ||
node.offsetParent === null;
return { hidden, text: (node.textContent || '').trim() };
}, el);
if (!info.hidden && info.text) {
shownErrors.push(info.text);
}
}
if (shownErrors.length > 0) {
console.log(`❌ Found ${shownErrors.length} visible error message(s) after a routine setting change`);
shownErrors.forEach((text, i) => console.log(` Error ${i + 1}: ${text}`));
// This is the whole point of the test — a visible error or
// validation message after a routine setting change is exactly
// what should fail it. It used to only log.
failed = true;
} else {
console.log(`✅ No visible error messages (${errorElements.length} empty/hidden error containers ignored)`);
}
} catch (error) {
console.error('❌ Test error:', error);
failed = true;
} finally {
await browser.close();
process.exit(failed ? 1 : 0);
}
}
testSettingsChange().catch(err => { console.error(err); process.exit(1); });