ulinycoin

[flagged]

há 4 mesesdiscuss

A few implementation details in case they're useful:

- LocalPDF runs fully in the browser — files are not uploaded to a server - The Studio UI is built as a canvas-style workspace with React Konva - Export/rebuild happens locally rather than through a server roundtrip - Studio is aimed at multi-document workflows, not just one-off conversions - I've been testing it on larger PDFs too, but performance still depends a lot on the user's machine because the whole point is local-first processing

The part I'm most curious about is whether the workspace/canvas model feels genuinely useful, or whether most people still prefer traditional tool-by-tool PDF UX.

ulinycoin

há 7 mesesdiscuss

I think indie products win on specific constraints.

If a user prompts generic stuff like "best pdf editor", the AI will likely route them to Adobe or the paid giants.

But users often prompt with constraints: "compress pdf locally", "convert pdf without uploading", or "pdf tools no signup".

That's where the indie product fits in. The big incumbents usually require uploads (for data harvesting) or logins (for growth). By strictly adhering to "privacy-first / local-only", my site satisfies a constraint that the big players structurally cannot.

The AI seems to recognize that distinction.

ulinycoin

há 7 mesesdiscuss

That's a very valid concern, especially for content-heavy sites (blogs, wikis). If an LLM can summarize the answer, the user has no reason to click.

However, I suspect there's a distinction between Information and Utilities.

An LLM can summarize a StackOverflow discussion on "how to compress a PDF," but it cannot (yet) reliably perform the heavy client-side processing to actually do it securely in the browser without uploading data.

For tools and utilities, the "click" is still necessary to perform the action. My bet is that AI will act more as a dispatcher for specific tasks ("Go here to fix X") rather than just a summarizer.

But you're right — once LLMs get native, sandboxed execution environments, even tools might get absorbed.

ulinycoin

há 7 mesesdiscuss

Adversarial branding is going to be a real niche in 2026.

ulinycoin

há 7 mesesdiscuss

Volume is modest (~180 visitors/month), but the 50/50 split is what's interesting.

Been in production since August 2025, so ~4 months.

The strategy was intentional from the start: there's no point competing with Adobe, Smallpdf, ILovePDF for Google rankings. They have 10+ years of backlinks, massive marketing budgets, and domain authority I'll never match as a solo dev.

So I made a bet on GEO from day one: - Semantic HTML that LLMs can parse - Clear technical docs (GitHub README as primary content) - Honest about limitations - Privacy-first architecture (client-side processing)

Your 5% GEO makes sense for a 2-year-old company optimizing for traditional SEO. The difference: I skipped the SEO game entirely. When you're competing in an established niche, GEO-first might be the only viable strategy for bootstrapped products.

Curious: what type of dev content are you creating? And have you tested how LLMs cite it vs your traditional marketing content?

ulinycoin

há 7 mesesdiscuss

Ha! AIO makes sense. Though knowing tech, we'll probably end up with 5 competing acronyms

ulinycoin

há 7 mesesdiscuss

Hi HN! I'm the creator of LocalPDF (https://localpdf.online) — a privacy-first PDF toolkit that works entirely in your browser.

I wanted to share something fascinating: ChatGPT has become my biggest traffic source, accounting for ~50% of visitors, compared to ~45% from Google. This happened organically — I didn't optimize specifically for AI assistants.

Key Stats: - ChatGPT: ~50% of traffic - Google Search: ~45% - Direct/Other: ~5% - Conversion rate from ChatGPT: 2x higher than Google - ChatGPT users stay 40% longer on average

What I think is happening:

GEO (Generative Engine Optimization) vs SEO Traditional SEO focuses on keywords, backlinks, and content clusters. But AI assistants work differently: - They parse semantic HTML structure - They value clear, descriptive content over keyword density - They prefer sites that solve problems directly

The ironic part? My site was built with basic semantic HTML and clear descriptions — not because I was targeting AI, but because it's good web development practice.

Technical Details: - Stack: Astro + TypeScript, client-side PDF.js - No tracking, no uploads — everything processes locally - Semantic HTML: proper headings, landmarks, meta descriptions - Clear tool descriptions with use cases

Why ChatGPT users convert better: 1. Intent clarity: They ask specific questions ("how to compress PDF without losing quality") 2. Contextual recommendations: ChatGPT explains why LocalPDF fits their needs 3. Trust transfer: AI assistant endorsement acts as social proof

Limitations: - Can't track which specific ChatGPT prompts drive traffic - No direct way to optimize for AI recommendations - ChatGPT's recommendations can change unpredictably

Questions for HN: 1. Are you seeing similar traffic patterns from AI assistants? 2. How do you measure/optimize for GEO vs traditional SEO? 3. What happens when every AI assistant has different ranking criteria?

This feels like early Google days — there's a level playing field where good products with clear value propositions can get discovered organically, without massive SEO budgets.

Happy to answer questions about the technical implementation, traffic patterns, or anything else!

ulinycoin

há 9 mesesdiscuss

Yes! LocalPDF is 100% client-side, so self-hosting is very simple.

  All PDF processing happens in the browser - no server needed. Just build
  and serve as static files:

  git clone https://github.com/ulinycoin/clientpdf-pro.git
  cd clientpdf-pro
  npm install && npm run build
  # Serve dist/ folder with any web server

  For Docker, you can use something like:
  FROM nginx:alpine
  COPY dist/ /usr/share/nginx/html
  EXPOSE 80

  No databases, no backend APIs, no file uploads to servers. It's just
  HTML/JS/CSS that runs entirely in your browser for privacy.

  The only "server" component in the repo is optional Rendertron for SEO
  (prerendering for search bots), but the core PDF tools work without it.

ulinycoin

há 9 mesesdiscuss

LocalPDF is a privacy-first set of PDF tools that run entirely in your browser. No files ever leave your device: all processing is done locally using modern web technologies, so you never have to worry about uploading sensitive documents to third-party servers.

Key features: - Merge, split, compress, and convert PDFs instantly—without server roundtrips - Edit your PDFs securely and privately, directly in your browser - 100% free and anonymous, with no sign-up required - Works on any operating system with a modern browser - Open source and transparent codebase

Perfect for users who need fast, no-compromise PDF processing while keeping their data secure and private. Try it now: https://localpdf.online

Submissions

Show HN: Shadow Web – Cut 64–97% of web page tokens for LLM agents

Show HN: Studio – a drag-and-drop PDF workspace for multi-document workflows

Show HN: I got 50% of my traffic from ChatGPT instead of Google

LocalPDF – Privacy-first PDF tools that work in browser

comments