living-os-cockpit

repository

loading code, commits, and activity

repositories

loading repo index

#1	import { NextRequest, NextResponse } from 'next/server';
#2	import { authErrorResponse, getUserContext } from '@/lib/user-context';
#3	import { writeFile, mkdir } from 'fs/promises';
#4	import { join } from 'path';
#5	import { request as httpRequest } from 'http';
#6	import { request as httpsRequest } from 'https';
#7
#8	export const runtime = 'nodejs';
#9	export const maxDuration = 600;
#10
#11	const MAX_PREVIEW_CHARS = 60000;
#12	const PDF_TEXT_LAYER_MIN_CHARS = 20;
#13
#14	function meaningfulChars(text: string) {
#15	return (text \|\| '').replace(/\s+/g, '').length;
#16	}
#17
#18	function postGatewayTool(token: string, payload: any, timeoutMs = 580000): Promise<{ status: number; data: any; text: string }> {
#19	return new Promise((resolve, reject) => {
#20	const target = new URL(`${process.env.GATEWAY_URL}/api/v1/tools/execute`);
#21	const body = JSON.stringify(payload);
#22	const requestImpl = target.protocol === 'https:' ? httpsRequest : httpRequest;
#23
#24	const req = requestImpl({
#25	hostname: target.hostname,
#26	port: target.port,
#27	path: `${target.pathname}${target.search}`,
#28	method: 'POST',
#29	headers: {
#30	'Authorization': `Bearer ${token}`,
#31	'Content-Type': 'application/json',
#32	'Content-Length': Buffer.byteLength(body),
#33	},
#34	}, (res) => {
#35	let text = '';
#36	res.setEncoding('utf8');
#37	res.on('data', (chunk) => { text += chunk; });
#38	res.on('end', () => {
#39	let data: any = null;
#40	try { data = JSON.parse(text); } catch {}
#41	resolve({ status: res.statusCode ?? 0, data, text });
#42	});
#43	});
#44
#45	req.setTimeout(timeoutMs, () => {
#46	req.destroy(new Error(`Gateway tool call timed out after ${timeoutMs}ms`));
#47	});
#48	req.on('error', reject);
#49	req.write(body);
#50	req.end();
#51	});
#52	}
#53
#54	export async function POST(req: NextRequest) {
#55	try {
#56	const ctx = await getUserContext();
#57	const formData = await req.formData();
#58	const file = formData.get('file') as File \| null;
#59
#60	if (!file) {
#61	return NextResponse.json({ error: 'No file provided' }, { status: 400 });
#62	}
#63
#64	const uploadDir = join(ctx.vaultPath, 'pending_uploads');
#65	await mkdir(uploadDir, { recursive: true });
#66
#67	const safeName = `${Date.now()}_${file.name.replace(/[^a-zA-Z0-9._-]/g, '_')}`;
#68	const filePath = join(uploadDir, safeName);
#69
#70	const buffer = Buffer.from(await file.arrayBuffer());
#71	await writeFile(filePath, buffer);
#72
#73	let preview = '';
#74	let extractionMethod: string \| undefined;
#75	let extractionPages: number \| undefined;
#76	let extractionWarning: string \| undefined;
#77	let truncated = false;
#78	let fullLength = 0;
#79
#80	const lower = safeName.toLowerCase();
#81	const isPdf = file.type === 'application/pdf' \|\| lower.endsWith('.pdf');
#82	const isText = file.type.startsWith('text/') \|\| /\.(txt\|md\|json\|csv\|log\|jsonl)$/i.test(safeName);
#83
#84	if (isPdf) {
#85	try {
#86	const textLayerCheck = await postGatewayTool(ctx.token, {
#87	tool: 'ocr_parse',
#88	args: { file_path: filePath, engine: 'pdfplumber' },
#89	},
#90	120000,
#91	);
#92
#93	if (textLayerCheck.status < 200 \|\| textLayerCheck.status >= 300) {
#94	extractionMethod = 'failed';
#95	const detail = typeof textLayerCheck.data?.detail === 'string'
#96	? textLayerCheck.data.detail
#97	: (textLayerCheck.text \|\| JSON.stringify(textLayerCheck.data)).slice(0, 200);
#98	extractionWarning = `Gateway OCR HTTP ${textLayerCheck.status}: ${detail}`;
#99	return NextResponse.json({
#100	filename: safeName,
#101	path: filePath,
#102	size: buffer.length,
#103	type: file.type,
#104	preview,
#105	extractionMethod,
#106	extractionPages,
#107	extractionWarning,
#108	truncated,
#109	fullLength,
#110	});
#111	}
#112
#113	const textResult = textLayerCheck.data?.result ?? textLayerCheck.data;
#114	const textLayerContent = textResult?.content ?? '';
#115	let result = textResult;
#116
#117	if (!textResult?.success \|\| meaningfulChars(textLayerContent) <= PDF_TEXT_LAYER_MIN_CHARS) {
#118	const chandraResult = await postGatewayTool(ctx.token, {
#119	tool: 'ocr_parse',
#120	args: { file_path: filePath, engine: 'chandra' },
#121	},
#122	580000,
#123	);
#124
#125	if (chandraResult.status < 200 \|\| chandraResult.status >= 300) {
#126	extractionMethod = 'failed';
#127	const detail = typeof chandraResult.data?.detail === 'string'
#128	? chandraResult.data.detail
#129	: (chandraResult.text \|\| JSON.stringify(chandraResult.data)).slice(0, 200);
#130	extractionWarning = `Gateway Chandra OCR HTTP ${chandraResult.status}: ${detail}`;
#131	return NextResponse.json({
#132	filename: safeName,
#133	path: filePath,
#134	size: buffer.length,
#135	type: file.type,
#136	preview,
#137	extractionMethod,
#138	extractionPages,
#139	extractionWarning,
#140	truncated,
#141	fullLength,
#142	});
#143	}
#144
#145	result = chandraResult.data?.result ?? chandraResult.data;
#146	if (textResult?.error && result && typeof result === 'object') {
#147	result.pdfplumber_error = textResult.error;
#148	}
#149	}
#150
#151	if (result?.success) {
#152	preview = result.content ?? '';
#153	fullLength = preview.length;
#154	extractionMethod = result.engine_used ?? 'chandra';
#155	extractionPages = result.page_count;
#156	if (preview.length > MAX_PREVIEW_CHARS) {
#157	preview = preview.slice(0, MAX_PREVIEW_CHARS) +
#158	`\n\n[... truncated. Full length: ${fullLength} chars. File at ${filePath}]`;
#159	truncated = true;
#160	}
#161
#162	// Fire-and-forget vault ingestion (non-blocking)
#163	if (result.content && fullLength > 100) {
#164	postGatewayTool(ctx.token, {
#165	tool: 'vault_ingest',
#166	args: {
#167	file_path: filePath,
#168	extracted_text: result.content,
#169	user_id: ctx.vaultUserKey,
#170	canonical_member_id: ctx.canonicalMemberId,
#171	},
#172	},
#173	600000,
#174	).catch((e) => console.error('vault_ingest fire-and-forget failed:', e?.message));
#175	}
#176	} else {
#177	extractionMethod = 'failed';
#178	extractionWarning = result?.error?.slice(0, 200) ?? 'OCR parse returned no content';
#179	}
#180	} catch (e: any) {
#181	extractionMethod = 'failed';
#182	extractionWarning = `Gateway tool call failed: ${e?.message?.slice(0, 200) ?? 'unknown'}`;
#183	}
#184	} else if (isText) {
#185	preview = buffer.toString('utf-8');
#186	fullLength = preview.length;
#187	if (preview.length > MAX_PREVIEW_CHARS) {
#188	preview = preview.slice(0, MAX_PREVIEW_CHARS) + `\n\n[... truncated]`;
#189	truncated = true;
#190	}
#191	extractionMethod = 'native';
#192	}
#193
#194	return NextResponse.json({
#195	filename: safeName,
#196	path: filePath,
#197	size: buffer.length,
#198	type: file.type,
#199	preview,
#200	extractionMethod,
#201	extractionPages,
#202	extractionWarning,
#203	truncated,
#204	fullLength,
#205	});
#206	} catch (error) {
#207	return authErrorResponse(error);
#208	}
#209	}
#210

z6Mku78K/living-os-cockpit