Files
dobetternorge-tools/assets/js/deep-research.js
T
daveadmin 4cbe0a4ac4 Add Deep Research tool — agent + rank/rerank RAG
New surface at /deep-research.php where the user pastes a question or
uploads PDF/DOCX/TXT case files and a LLM-orchestrated agent researches
the Do Better Norge legal corpus from 3-5 angles, with hybrid retrieval,
cross-encoder rerank, and synthesis that emits an inline-[n]-cited
markdown brief plus a numbered sources panel.

Uploaded documents are chunked + embedded in memory only (nomic-embed-text
via LiteLLM) and searched alongside the shared corpus during the same
request — never persisted to disk, DB, or Qdrant.

Reuses ClientRagPipeline::searchAll (hybrid + rerank), dbnV6 slice
helpers, and the existing extract.php text-extraction logic via a new
dbnToolsExtractUploadedFile() helper. Also adds dbnToolsCallGpuLlm()
helper in bootstrap.php — fixes a latent bug where LegalTools.php
was already calling that name with no definition.

Search.php is unchanged.
2026-05-15 10:30:47 +02:00

482 lines
18 KiB
JavaScript
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
/* deep-research.js — page-scoped UI for /deep-research.php */
(function () {
'use strict';
const els = {};
let lang = 'en';
let uploadFiles = [];
let lastResult = null;
const SLICE_DEFS = [
{ id: 'family_core', label: 'Family Law Core' },
{ id: 'child_welfare', label: 'Child Welfare' },
{ id: 'echr_hague', label: 'ECHR and Hague' },
{ id: 'broader_legal', label: 'Broader Legal Support' },
];
const STEP_LABELS = [
'Query interpretation',
'Query expansion',
'Slice resolution',
'Upload indexing',
'Retrieval',
'Synthesis',
'Citation confidence',
];
document.addEventListener('DOMContentLoaded', () => {
if (!document.body.dataset.activeTool || document.body.dataset.activeTool !== 'deep-research') return;
Object.assign(els, {
form: document.getElementById('deepResearchForm'),
input: document.getElementById('drInput'),
status: document.getElementById('drStatus'),
runButton: document.getElementById('drRunButton'),
results: document.getElementById('drResults'),
traceList: document.getElementById('traceList'),
slices: Array.from(document.querySelectorAll('.dr-slice')),
langButtons: Array.from(document.querySelectorAll('#drLangSwitcher .lang-btn')),
engineRadios: Array.from(document.querySelectorAll('input[name="drEngine"]')),
subQ: document.getElementById('drSubQ'),
subQVal: document.getElementById('drSubQValue'),
chunkLimit: document.getElementById('drChunkLimit'),
chunkLimitVal: document.getElementById('drChunkLimitValue'),
sim: document.getElementById('drSim'),
simVal: document.getElementById('drSimValue'),
topK: document.getElementById('drTopK'),
topKVal: document.getElementById('drTopKValue'),
temp: document.getElementById('drTemp'),
tempVal: document.getElementById('drTempValue'),
uploadZone: document.getElementById('drUploadZone'),
uploadInput: document.getElementById('drUploadInput'),
uploadPrompt: document.getElementById('drUploadPrompt'),
uploadFileInfo: document.getElementById('drUploadFileInfo'),
uploadFileList: document.getElementById('drUploadFileList'),
uploadClear: document.getElementById('drUploadClear'),
modal: document.getElementById('drSourceModal'),
modalClose: document.getElementById('drSourceModalClose'),
modalTitle: document.getElementById('drSourceModalTitle'),
modalEyebrow: document.getElementById('drSourceModalEyebrow'),
modalMeta: document.getElementById('drSourceModalMeta'),
modalText: document.getElementById('drSourceModalText'),
});
if (!els.form) return;
bindSlices();
bindLang();
bindRanges();
bindUpload();
bindModal();
els.form.addEventListener('submit', onSubmit);
// Pre-render placeholder trace
renderTrace(STEP_LABELS.map((label) => ({ label, detail: 'Waiting…', status: 'idle' })));
});
function bindSlices() {
els.slices.forEach((btn) => {
btn.addEventListener('click', () => {
const isOn = btn.classList.toggle('is-on');
btn.setAttribute('aria-pressed', isOn ? 'true' : 'false');
const badge = btn.querySelector('.dr-slice__badge');
if (badge) badge.textContent = isOn ? 'on' : 'off';
});
});
}
function bindLang() {
els.langButtons.forEach((b) => {
b.addEventListener('click', () => {
els.langButtons.forEach((x) => x.classList.remove('is-active'));
b.classList.add('is-active');
lang = b.dataset.lang || 'en';
});
});
}
function bindRanges() {
const pairs = [
[els.subQ, els.subQVal, (v) => v],
[els.chunkLimit, els.chunkLimitVal, (v) => v],
[els.sim, els.simVal, (v) => Number(v).toFixed(2)],
[els.topK, els.topKVal, (v) => v],
[els.temp, els.tempVal, (v) => Number(v).toFixed(2)],
];
pairs.forEach(([range, label, fmt]) => {
if (!range || !label) return;
const sync = () => { label.textContent = fmt(range.value); };
range.addEventListener('input', sync);
sync();
});
}
function bindUpload() {
if (!els.uploadZone) return;
const onFiles = (fileList) => {
const files = Array.from(fileList || []).slice(0, 5);
if (uploadFiles.length + files.length > 5) {
setStatus('At most 5 files can be uploaded per request.', 'error');
return;
}
files.forEach((f) => {
if (f.size > 4 * 1024 * 1024) {
setStatus(`${f.name} exceeds the 4 MB limit.`, 'error');
return;
}
const ext = (f.name.split('.').pop() || '').toLowerCase();
if (!['pdf', 'docx', 'txt'].includes(ext)) {
setStatus(`${f.name} is not a supported file type.`, 'error');
return;
}
uploadFiles.push(f);
});
renderUploadList();
};
els.uploadInput.addEventListener('change', (e) => onFiles(e.target.files));
els.uploadZone.addEventListener('dragover', (e) => { e.preventDefault(); els.uploadZone.classList.add('is-drop'); });
els.uploadZone.addEventListener('dragleave', () => els.uploadZone.classList.remove('is-drop'));
els.uploadZone.addEventListener('drop', (e) => {
e.preventDefault();
els.uploadZone.classList.remove('is-drop');
onFiles(e.dataTransfer?.files);
});
els.uploadClear?.addEventListener('click', () => {
uploadFiles = [];
els.uploadInput.value = '';
renderUploadList();
});
}
function renderUploadList() {
if (!uploadFiles.length) {
els.uploadFileInfo.classList.add('is-hidden');
els.uploadPrompt.classList.remove('is-hidden');
return;
}
els.uploadPrompt.classList.add('is-hidden');
els.uploadFileInfo.classList.remove('is-hidden');
els.uploadFileList.innerHTML = uploadFiles.map((f, i) => {
const kb = (f.size / 1024).toFixed(0);
return `<li><span class="upload-filename">${escapeHtml(f.name)}</span><span class="upload-chars">${kb} KB</span></li>`;
}).join('');
}
function bindModal() {
els.modalClose?.addEventListener('click', closeModal);
els.modal?.addEventListener('click', (e) => {
if (e.target === els.modal) closeModal();
});
document.addEventListener('keydown', (e) => {
if (e.key === 'Escape' && els.modal && !els.modal.classList.contains('is-hidden')) closeModal();
});
}
function closeModal() {
els.modal?.classList.add('is-hidden');
}
function openModal(source) {
if (!source) return;
els.modalEyebrow.textContent = source.source_origin === 'upload' ? 'Uploaded file' : 'Corpus source';
els.modalTitle.textContent = source.title || 'Source';
const metaRows = [
['Number', `[${source.n}]`],
source.section ? ['Section', source.section] : null,
['Corpus / package', source.package_or_corpus || '—'],
source.authority_type ? ['Authority', source.authority_type] : null,
source.jurisdiction ? ['Jurisdiction', source.jurisdiction] : null,
source.similarity != null ? ['Similarity', String(source.similarity)] : null,
source.reranker_score != null ? ['Rerank score', String(source.reranker_score)] : null,
source.matched_sub_questions?.length ? ['Matched sub-Q', source.matched_sub_questions.join(', ')] : null,
].filter(Boolean);
els.modalMeta.innerHTML = '<dl>' + metaRows.map(([k, v]) => `<dt>${escapeHtml(k)}</dt><dd>${escapeHtml(String(v))}</dd>`).join('') + '</dl>';
els.modalText.textContent = source.chunk_text || source.excerpt || '';
els.modal.classList.remove('is-hidden');
}
function getSelectedSlices() {
const out = {};
SLICE_DEFS.forEach((s) => {
const btn = els.slices.find((b) => b.dataset.slice === s.id);
out[s.id] = !!(btn && btn.classList.contains('is-on'));
});
return out;
}
function getEngine() {
const checked = els.engineRadios.find((r) => r.checked);
return checked ? checked.value : 'azure_mini';
}
function getControls() {
return {
sub_q_count: parseInt(els.subQ.value, 10),
chunk_limit: parseInt(els.chunkLimit.value, 10),
similarity_threshold: parseFloat(els.sim.value),
reranker_top_k: parseInt(els.topK.value, 10),
temperature: parseFloat(els.temp.value),
};
}
async function onSubmit(e) {
e.preventDefault();
const query = (els.input.value || '').trim();
if (!query && uploadFiles.length === 0) {
setStatus('Type a question or upload a file before running deep research.', 'error');
return;
}
const slices = getSelectedSlices();
if (!Object.values(slices).some(Boolean)) {
setStatus('Enable at least one corpus slice.', 'error');
return;
}
setStatus('Running deep research…', 'busy');
els.runButton.disabled = true;
els.results.innerHTML = `<div class="empty-state"><h3>Working…</h3><p>The agent is expanding the question, retrieving from the corpus, and synthesising the brief. This usually takes 615 seconds.</p></div>`;
// Render placeholder trace with first step running
const placeholder = STEP_LABELS.map((label, i) => ({
label,
detail: i === 0 ? 'Running…' : 'Queued',
status: i === 0 ? 'running' : 'idle',
}));
renderTrace(placeholder);
const payload = {
query,
paste_text: '',
slices,
engine: getEngine(),
language: lang,
controls: getControls(),
};
let response;
try {
if (uploadFiles.length > 0) {
const form = new FormData();
form.append('payload', JSON.stringify(payload));
uploadFiles.forEach((f) => form.append('files[]', f));
response = await fetch('api/deep-research.php', { method: 'POST', body: form, credentials: 'same-origin' });
} else {
response = await fetch('api/deep-research.php', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify(payload),
credentials: 'same-origin',
});
}
} catch (err) {
setStatus(`Network error: ${err.message || err}`, 'error');
els.runButton.disabled = false;
return;
}
let data = null;
try { data = await response.json(); } catch (_) {}
if (!response.ok || !data || data.ok === false) {
const msg = (data && data.error && data.error.message) || `Request failed (${response.status}).`;
setStatus(msg, 'error');
els.runButton.disabled = false;
renderTrace(placeholder.map((s, i) => i === 0 ? { ...s, status: 'error', detail: msg } : s));
return;
}
lastResult = data;
setStatus(`Done in ${data.latency_ms || 0} ms · ${data.trace_metadata?.source_count || 0} sources · confidence ${data.trace_metadata?.citation_confidence || '?'}`, 'ok');
els.runButton.disabled = false;
renderTrace(data.trace || []);
renderResults(data);
}
function setStatus(message, kind) {
els.status.textContent = message;
els.status.style.color = kind === 'error' ? '#b41e1e' : (kind === 'ok' ? 'var(--teal-dark)' : 'var(--muted)');
}
function renderTrace(steps) {
if (!els.traceList) return;
els.traceList.classList.add('is-rich');
els.traceList.innerHTML = steps.map((step, i) => {
const statusClass = step.status === 'running' ? 'is-running'
: step.status === 'complete' ? 'is-done'
: step.status === 'warning' ? 'is-warning'
: step.status === 'error' ? 'is-error'
: '';
const marker = step.status === 'complete' ? '✓'
: step.status === 'warning' ? '!'
: step.status === 'error' ? '×'
: (i + 1);
return `<li class="trace-step ${statusClass}">
<span class="trace-step__marker">${marker}</span>
<div>
<span class="trace-step__label">${escapeHtml(step.label || '')}</span>
<span class="trace-step__detail">${escapeHtml(step.detail || '')}</span>
</div>
</li>`;
}).join('');
}
function renderResults(data) {
const sources = data.sources || [];
const subs = data.sub_questions || [];
const briefHtml = renderBrief(data.brief_markdown || '', sources);
const subQHtml = subs.length ? `
<div class="dr-result-block">
<h3 style="margin:0 0 8px;font-size:1rem">Angles the agent explored</h3>
<ol style="padding-left:1.2em;margin:0;color:var(--muted);line-height:1.55">
${subs.map((sq) => `<li><strong style="color:var(--ink)">${escapeHtml(sq.question)}</strong>${sq.rationale ? `<br><small>${escapeHtml(sq.rationale)}</small>` : ''}</li>`).join('')}
</ol>
</div>` : '';
const sourcesHtml = `
<div class="dr-result-block">
<div class="dr-sources-head">
<h3>Sources (${sources.length})</h3>
<small>Click a card to see the full chunk + scores</small>
</div>
<div class="dr-source-list">
${sources.map((s) => renderSourceCard(s)).join('')}
</div>
</div>`;
const uncertHtml = (data.what_remains_uncertain || []).length ? `
<div class="dr-result-block">
<h3 style="margin:0 0 8px;font-size:0.95rem;color:var(--muted)">What remains uncertain</h3>
<ul style="padding-left:1.2em;margin:0;color:var(--muted);line-height:1.55">
${(data.what_remains_uncertain || []).map((u) => `<li>${escapeHtml(String(u))}</li>`).join('')}
</ul>
</div>` : '';
const nextHtml = data.next_practical_step ? `
<div class="dr-result-block">
<h3 style="margin:0 0 6px;font-size:0.95rem">Next practical step</h3>
<p style="margin:0;color:var(--ink);line-height:1.5">${escapeHtml(data.next_practical_step)}</p>
</div>` : '';
els.results.innerHTML = `
<div class="dr-result-block">
<div class="dr-brief">${briefHtml}</div>
</div>
${subQHtml}
${sourcesHtml}
${uncertHtml}
${nextHtml}
`;
// Bind source-card click handlers + citation marker click handlers
els.results.querySelectorAll('[data-source-n]').forEach((node) => {
node.addEventListener('click', () => {
const n = parseInt(node.dataset.sourceN, 10);
const src = sources.find((s) => s.n === n);
if (src) {
openModal(src);
flashSource(n);
}
});
});
}
function flashSource(n) {
document.querySelectorAll('.dr-source-card.is-highlight').forEach((c) => c.classList.remove('is-highlight'));
const target = document.querySelector(`.dr-source-card[data-source-n="${n}"]`);
if (target) {
target.classList.add('is-highlight');
target.scrollIntoView({ behavior: 'smooth', block: 'center' });
setTimeout(() => target.classList.remove('is-highlight'), 1800);
}
}
function renderSourceCard(s) {
const score = s.reranker_score != null ? s.reranker_score : s.similarity;
const originTagClass = s.source_origin === 'upload' ? 'dr-source-tag dr-source-tag--upload' : 'dr-source-tag';
const originLabel = s.source_origin === 'upload' ? 'upload' : 'corpus';
return `<button type="button" class="dr-source-card" data-source-n="${s.n}">
<span class="dr-source-number">${s.n}</span>
<div class="dr-source-body">
<div class="dr-source-title">${escapeHtml(s.title || 'Untitled')}</div>
${s.section ? `<div class="dr-source-meta"><span class="dr-source-tag">${escapeHtml(s.section)}</span></div>` : ''}
<div class="dr-source-meta">
<span class="${originTagClass}">${originLabel}</span>
<span class="dr-source-tag dr-source-tag--score">${escapeHtml(s.package_or_corpus || '—')}</span>
${(s.matched_sub_questions || []).map((q) => `<span class="dr-source-tag">${escapeHtml(q)}</span>`).join('')}
</div>
<p class="dr-source-excerpt">${escapeHtml(truncate(s.excerpt || '', 240))}</p>
</div>
<div class="dr-source-aside">
<span>score<br><b>${score != null ? Number(score).toFixed(2) : '—'}</b></span>
${s.reranker_score != null && s.similarity != null ? `<span>sim<br><b>${Number(s.similarity).toFixed(2)}</b></span>` : ''}
</div>
</button>`;
}
// Markdown renderer — minimal: paragraphs, bold/italic, code, [n] citation badges
function renderBrief(markdown, sources) {
if (!markdown) return '<p><em>No brief was returned.</em></p>';
const sourceSet = new Set((sources || []).map((s) => s.n));
const escaped = escapeHtml(markdown);
// Citation markers [1], [1,2], [1-3]
const withCites = escaped.replace(/\[(\d+(?:\s*[-,]\s*\d+)*)\]/g, (_, group) => {
const nums = expandCiteGroup(group);
return nums.map((n) => {
const known = sourceSet.has(n);
const cls = known ? 'dr-cite' : 'dr-cite';
return `<span class="${cls}" data-source-n="${n}" role="button" tabindex="0">${n}</span>`;
}).join('');
});
// Bold/italic
const withBold = withCites
.replace(/\*\*([^*]+)\*\*/g, '<strong>$1</strong>')
.replace(/(^|[^*])\*([^*]+)\*(?!\*)/g, '$1<em>$2</em>')
.replace(/`([^`]+)`/g, '<code>$1</code>');
// Paragraphs
const paragraphs = withBold.split(/\n{2,}/).map((p) => {
const t = p.trim();
if (!t) return '';
if (/^### /.test(t)) return `<h4 style="margin:14px 0 6px;color:var(--ink);font-size:1rem">${t.replace(/^### /, '')}</h4>`;
return `<p>${t.replace(/\n/g, '<br>')}</p>`;
}).join('');
return paragraphs;
}
function expandCiteGroup(group) {
const out = [];
group.split(',').forEach((part) => {
const range = part.trim().match(/^(\d+)\s*-\s*(\d+)$/);
if (range) {
const a = parseInt(range[1], 10);
const b = parseInt(range[2], 10);
for (let i = a; i <= b; i++) out.push(i);
} else {
const n = parseInt(part.trim(), 10);
if (!Number.isNaN(n)) out.push(n);
}
});
return Array.from(new Set(out));
}
function escapeHtml(s) {
return String(s)
.replace(/&/g, '&amp;')
.replace(/</g, '&lt;')
.replace(/>/g, '&gt;')
.replace(/"/g, '&quot;')
.replace(/'/g, '&#039;');
}
function truncate(s, n) {
if (!s) return '';
if (s.length <= n) return s;
return s.slice(0, n - 1) + '…';
}
})();