Make any PDF readable for AI

# 1. PDF → Markdown. Inline the file as a base64 data URI (10 MB max per item).
curl https://inference.aivax.net/api/v1/web/fetch \
  -H "Authorization: Bearer $AIVAX_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{
    \"contents\": [\"data:application/pdf;base64,$(base64 -w0 manual.pdf)\"],
    \"returnErrors\": true
  }"

# → { "data": { "results": [ {
#       "index": 0,
#       "extractedText": "# Title\n\nParagraph...",   ← Markdown
#       "processingUnits": 7,                      ← what you pay for
#       "jsonProcessingUnits": 0,
#       "error": null } ] } }

# 2. Markdown → segments. `sanitize` skips fragments that are useless for retrieval.
curl https://inference.aivax.net/api/v1/generations/segment \
  -H "Authorization: Bearer $AIVAX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "documents": ["# Title\n\nParagraph..."], "sanitize": true }'

# → { "data": {
#       "result": [ { "index": 0, "count": 8, "segments": ["...", "..."] } ],
#       "usage":  { "processing_units": 2516, "cost": 0 } } }

// Works in Node 18+, Bun, Deno and browsers. Keep the API key server-side.
const AIVAX = "https://inference.aivax.net/api/v1";
const headers = {
  authorization: `Bearer ${process.env.AIVAX_API_KEY}`,
  "content-type": "application/json",
};

// 1. PDF → Markdown
async function pdfToMarkdown(bytes) {
  const base64 = Buffer.from(bytes).toString("base64");

  const response = await fetch(`${AIVAX}/web/fetch`, {
    method: "POST",
    headers,
    body: JSON.stringify({
      contents: [`data:application/pdf;base64,${base64}`],
      returnErrors: true,
    }),
  });

  const { data } = await response.json();
  const [item] = data.results;          // one result per `contents` entry, matched by index

  if (item.error) throw new Error(item.error);

  return { markdown: item.extractedText, processingUnits: item.processingUnits };
}

// 2. Markdown → RAG segments
async function segmentForRag(markdown, { sanitize = true } = {}) {
  const response = await fetch(`${AIVAX}/generations/segment`, {
    method: "POST",
    headers,
    body: JSON.stringify({ documents: [markdown], sanitize }),
  });

  const { data } = await response.json();

  return { segments: data.result[0].segments, processingUnits: data.usage.processing_units };
}

// 3. Shape the segments as JSONL for POST /api/v1/collections/{id}/documents
function toJsonl(segments, source) {
  return segments
    .map((text, index) => JSON.stringify({
      docid: `${source}#${index + 1}`,       // stable name: re-importing updates instead of duplicating
      text,
      __ref: source.slice(0, 64),           // groups the chunks of one file
      __tags: ["pdf"],
      __meta: { source, index },
    }))
    .join("\n");
}

const bytes = await Bun.file("manual.pdf").arrayBuffer();
const { markdown } = await pdfToMarkdown(bytes);
const { segments } = await segmentForRag(markdown);

await Bun.write("manual.jsonl", toJsonl(segments, "manual.pdf"));

// Trimmed from src/worker.js in the repository. The real file adds Turnstile
// verification and a Durable Object sliding-window rate limiter.
const AIVAX_API = "https://inference.aivax.net/api/v1";

export default {
  async fetch(request, env) {
    const url = new URL(request.url);

    if (url.pathname === "/api/convert" && request.method === "POST") {
      const form = await request.formData();
      const file = form.get("file");

      const gate = await verifyTurnstileAndQuota(request, env, form.get("turnstile"));
      if (gate instanceof Response) return gate;      // 403 or 429

      const base64 = bytesToBase64(new Uint8Array(await file.arrayBuffer()));
      const call = await callAivax(env, "/web/fetch", {
        contents: [`data:application/pdf;base64,${base64}`],
        returnErrors: true,
      });

      const item = call.data.results[0];
      return Response.json({
        markdown: item.extractedText,
        processingUnits: item.processingUnits,
        elapsedMs: call.elapsedMs,
        requestId: call.requestId,
      });
    }

    return env.ASSETS.fetch(request);                 // static site from ./public
  },
};

async function callAivax(env, path, body) {
  const startedAt = Date.now();
  const response = await fetch(AIVAX_API + path, {
    method: "POST",
    headers: {
      authorization: `Bearer ${env.AIVAX_API_KEY}`,       // `wrangler secret put AIVAX_API_KEY`
      "content-type": "application/json",
    },
    body: JSON.stringify(body),
  });

  const payload = await response.json();

  return {
    data: payload.data,
    elapsedMs: Date.now() - startedAt,
    requestId: response.headers.get("x-request-id"),
  };
}