Client-side PDF Q&A

Try it live → Run the PDF Q&A demo — drop a PDF, ask questions, all in your browser.

The shape

You have a doc your team needs to chat with. You don’t want to build a RAG pipeline. You don’t want to upload PDFs to a vendor. You want the fastest path from “drop file” to “ask question.”

<script src="https://unpkg.com/@linda/script/dist/linda.js"></script>
<script type="module">
  // Lazy-register the PDF parser on linda:ready (fires after auto-init).
  window.addEventListener("linda:ready", async (e) => {
    const linda = e.detail;
    const { pdfTextParser } = await import("https://esm.sh/@linda/parsers-pdf");
    linda.parsers.add(pdfTextParser);
  });
</script>

<form
  data-linda
  data-linda-config="/configs/pdf-qna.json"
  data-linda-provider="anthropic"
  data-linda-key="sk-ant-..."
>
  <input name="pdf" type="file" accept="application/pdf" />
  <textarea name="question"></textarea>
</form>

That’s it. Drop a PDF, ask “what’s the indemnity cap?”, get a citation.

What’s happening

1. PDF dropped onto the page → onFileUpload hook fires.
2. @linda/parsers-pdf (pdf.js) parses the file in-browser. Output lands in /user/files/<id>/parsed/text.md and pages.jsonl.
3. User asks a question. The agent reads the parsed text and answers, quoting the source if relevant.
4. The raw PDF stays in browser memory. Nothing uploaded.

Compliance angle

• GDPR: Linda doesn’t process personal data on a server. There’s no controller-processor relationship to model.
• HIPAA-friendly: If your provider has a BAA, you only need to worry about the prompts that hit it — the file itself is local.
• No vendor lock: Swap providers without re-uploading documents anywhere.

Cost

Per Q: ~5–10 K tokens of context (the parsed PDF chunks the model picks) + 500-token answer. **$0.02 per question** on Sonnet 4.6, with prompt caching making follow-up Qs ~$0.002.

FAQ