~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 byte-size formatter.
*
* Single source of truth for "1536 -> 1.5 KB" style formatting, previously
* duplicated as formatBytes in deletion/delete_manager.js and formatFileSize
* in pdf_upload_handler.js. Loaded globally from base.html so both consumers
* (and any future ones) can call window.formatBytes.
*/
(function() {
'use strict';
/**
* Format a byte count as a human-readable size.
* @param {number} bytes - Non-negative byte count.
* @returns {string} e.g. "0 Bytes", "1.5 KB", "1 MB".
*/
function formatBytes(bytes) {
// Covers 0, negatives, NaN, undefined, and null. Relational
// comparison coerces null to 0 and undefined to NaN.
// Without it, Math.log(-1) is NaN, which propagates through the
// clamp and yields "NaN undefined". Number.isFinite additionally
// rejects Infinity, which the relational check lets through and
// the unit clamp would render as "Infinity EB".
if (!(bytes > 0) || !Number.isFinite(bytes)) return '0 Bytes';
const k = 1024;
const sizes = ['Bytes', 'KB', 'MB', 'GB', 'TB', 'PB', 'EB'];
// Clamp both ends: the upper bound stops indexing past the unit list
// for very large values, and the lower bound keeps sub-1-byte inputs
// (log < 0) from producing a negative index -> undefined unit.
const i = Math.max(0, Math.min(Math.floor(Math.log(bytes) / Math.log(k)), sizes.length - 1));
return parseFloat((bytes / k ** i).toFixed(2)) + ' ' + sizes[i];
}
window.formatBytes = formatBytes;
})();