acmcc/supabase/functions/parse-invoice/index.ts

const corsHeaders = {
  "Access-Control-Allow-Origin": "*",
  "Access-Control-Allow-Headers":
    "authorization, x-client-info, apikey, content-type, x-supabase-client-platform, x-supabase-client-platform-version, x-supabase-client-runtime, x-supabase-client-runtime-version",
};

function sanitizeJsonCandidate(value: string) {
  return value
    .replace(/^```(?:json)?\s*/i, "")
    .replace(/```$/i, "")
    .replace(/[\u0000-\u001F\u007F-\u009F]/g, (char) =>
      char === "\n" || char === "\r" || char === "\t" ? char : " ",
    )
    .replace(/,\s*([}\]])/g, "$1")
    .trim();
}

function extractJsonObject(value: string) {
  const fenced = value.match(/```(?:json)?\s*([\s\S]*?)```/i)?.[1];
  const source = fenced || value;
  const start = source.indexOf("{");
  if (start === -1) return sanitizeJsonCandidate(source);

  let depth = 0;
  let inString = false;
  let escaped = false;
  for (let i = start; i < source.length; i++) {
    const char = source[i];
    if (escaped) {
      escaped = false;
      continue;
    }
    if (char === "\\") {
      escaped = true;
      continue;
    }
    if (char === '"') {
      inString = !inString;
      continue;
    }
    if (inString) continue;
    if (char === "{") depth++;
    if (char === "}") depth--;
    if (depth === 0) return sanitizeJsonCandidate(source.slice(start, i + 1));
  }

  throw new Error("AI response was truncated before the JSON object closed");
}

function parseAiJson(content: string) {
  const candidates = [content.trim(), extractJsonObject(content)];
  for (const candidate of candidates) {
    try {
      return JSON.parse(sanitizeJsonCandidate(candidate));
    } catch {
      // try the next extraction strategy
    }
  }
  throw new Error("Failed to parse AI response as valid JSON");
}

Deno.serve(async (req) => {
  if (req.method === "OPTIONS") {
    return new Response(null, { headers: corsHeaders });
  }

  try {
    const { pdf_base64, filename } = await req.json();

    if (!pdf_base64) {
      return new Response(JSON.stringify({ error: "No PDF data provided" }), {
        status: 400,
        headers: { ...corsHeaders, "Content-Type": "application/json" },
      });
    }

    const LOVABLE_API_KEY = Deno.env.get("LOVABLE_API_KEY");
    if (!LOVABLE_API_KEY) {
      return new Response(JSON.stringify({ error: "AI API key not configured" }), {
        status: 500,
        headers: { ...corsHeaders, "Content-Type": "application/json" },
      });
    }

    // Guard against oversized/scanned PDFs that the AI gateway rejects with an
    // opaque error. The caller (inbound-bill-email) still keeps the attachment
    // in the inbox for manual entry when this returns an error.
    const approxBytes = Math.floor((pdf_base64.length * 3) / 4);
    const sizeMB = approxBytes / (1024 * 1024);
    const MAX_MB = 18;
    if (sizeMB > MAX_MB) {
      return new Response(
        JSON.stringify({ error: `PDF too large to auto-parse (${sizeMB.toFixed(1)} MB; limit ${MAX_MB} MB). Saved to the inbox for manual entry.` }),
        { status: 413, headers: { ...corsHeaders, "Content-Type": "application/json" } },
      );
    }

    const prompt = `You are a meticulous invoice data extraction AI. Analyze the provided PDF invoice carefully — examine EVERY page and EVERY line item, including continuation pages, sub-totals, and tables. Do not skip or summarize line items; capture each one individually exactly as it appears.

CRITICAL RULES:
- Read the entire document end-to-end before responding.
- Extract EVERY line item separately, even if there are 50+ rows. Do not collapse, group, or omit any.
- Preserve the exact wording from the invoice for descriptions and names.
- For dates, convert to YYYY-MM-DD. If only month/year is shown, use the 1st of the month.
- For monetary values, use numbers only (no currency symbols, no commas). Negative amounts (credits/discounts) should be negative numbers.
- If a field truly cannot be determined from the document, use null for strings and 0 for numbers — do NOT guess.
- The sum of line item amounts should reconcile with the subtotal; double-check before returning.
- Return a complete, syntactically valid JSON object with no markdown fences, comments, trailing commas, or extra text.
- The sum of line item amounts should reconcile with the subtotal; double-check before returning.

Return ONLY valid JSON (no markdown, no code blocks, no commentary) with this exact structure:
{
  "vendor_name": "string",
  "vendor_address": "string or null",
  "vendor_phone": "string or null",
  "client_name": "string or null",
  "client_address": "string or null",
  "invoice_number": "string",
  "invoice_date": "YYYY-MM-DD",
  "due_date": "YYYY-MM-DD or null",
  "service_period": "string or null",
  "subtotal": number,
  "tax": number,
  "other_charges": number,
  "total_amount": number,
  "currency": "USD",
  "payment_terms": "string or null",
  "notes": "string or null",
  "line_items": [
    {
      "line_number": number,
      "description": "string (full description as printed)",
      "name": "string (short item name/SKU if present, else first words of description)",
      "date": "YYYY-MM-DD or null",
      "quantity": number or null,
      "unit_price": number or null,
      "amount": number,
      "category": "string or null",
      "notes": "string or null"
    }
  ]
}`;

    const response = await fetch("https://ai.gateway.lovable.dev/v1/chat/completions", {
      method: "POST",
      headers: {
        "Content-Type": "application/json",
        Authorization: `Bearer ${LOVABLE_API_KEY}`,
      },
      body: JSON.stringify({
        model: "google/gemini-2.5-pro",
        messages: [
          { role: "system", content: prompt },
          {
            role: "user",
            content: [
              {
                type: "text",
                text: `Extract ALL invoice data from this PDF file named "${filename || "invoice.pdf"}". Examine every page and capture every line item — do not skip or summarize any rows. Return only the JSON object.`,
              },
              {
                type: "image_url",
                image_url: {
                  url: `data:application/pdf;base64,${pdf_base64}`,
                },
              },
            ],
          },
        ],
        temperature: 0,
        max_tokens: 32000,
        response_format: { type: "json_object" },
      }),
    });

    if (!response.ok) {
      const errText = await response.text();
      console.error("AI Gateway error:", response.status, errText);
      const snippet = errText.replace(/\s+/g, " ").slice(0, 300);
      const hint = response.status === 429 || response.status === 402
        ? " (AI gateway rate/credit limit)"
        : response.status === 413 || /too large|payload|size/i.test(errText)
        ? " (PDF too large for the AI gateway — saved to inbox for manual entry)"
        : "";
      return new Response(
        JSON.stringify({ error: `AI gateway error ${response.status}${hint}: ${snippet || "no detail"}` }),
        { status: 502, headers: { ...corsHeaders, "Content-Type": "application/json" } },
      );
    }

    const aiResult = await response.json();
    const content = aiResult.choices?.[0]?.message?.content || "";

    let parsed;
    try {
      parsed = parseAiJson(typeof content === "string" ? content : JSON.stringify(content));
    } catch (parseError) {
      console.error("Failed to parse AI response:", content);
      return new Response(JSON.stringify({ error: parseError.message || "Failed to parse AI response" }), {
        status: 422,
        headers: { ...corsHeaders, "Content-Type": "application/json" },
      });
    }

    return new Response(JSON.stringify({ data: parsed }), {
      headers: { ...corsHeaders, "Content-Type": "application/json" },
    });
  } catch (err) {
    console.error("parse-invoice error:", err);
    return new Response(JSON.stringify({ error: err.message }), {
      status: 500,
      headers: { ...corsHeaders, "Content-Type": "application/json" },
    });
  }
});