About Codeswap

Codeswap is a developer tools site. It's built and maintained by one person, runs on a single Hostinger server, and pays for itself with display ads. This page covers what's on the site, who built it, why it exists, and how to get in touch.

What's on Codeswap

Three things, in order of importance:

  1. Interactive developer tools. Token counters for every major LLM, JSON and YAML formatters, a JWT decoder, regex tester, hash and HMAC generators, base64 and URL encoders, a cURL command converter, and a growing list of others. Each tool runs entirely in your browser — input never leaves your device. The full catalog lives under the LLM, JSON, Regex, Crypto, and other category pages in the header navigation.
  2. A current LLM model database. Specs, pricing, context windows, modalities, and benchmark scores for 25+ frontier and open-weight models — OpenAI's GPT-5 line, the Claude 4.x series, Gemini 2.x, the Llama and Qwen open-weight families, DeepSeek, Mistral, Cohere, xAI. The database powers the model pages at /models/, side-by-side comparisons at /compare/, and use-case rankings at /best-llm-for-X/.
  3. A prompts library. A growing collection of prompts that have proven useful in production — system prompts, evaluation rubrics, RAG patterns, code-review prompts. Each prompt includes recommended models and example outputs. Users can submit prompts; submissions are moderated before publishing.

Who built it

Sathvic Kollu. I've been writing software for over a decade, with the last several years focused on LLM-powered applications. I'm based in India and reachable at [email protected].

Codeswap is not a venture-backed product. There's no team. The site does not employ "expert authors" or anonymous contributors. Every page was written and edited by me before going live. If a fact on a model page is wrong, the responsibility is mine — there is a "report inaccuracy" link on every page and an email above. I read and respond.

Why this site exists

A few practical reasons. First, the existing tools I used daily — for token counting, JSON formatting, JWT debugging — were either slow, ad-laden, or shipped wrong output for edge cases. I built versions that solved my own problems first, then extended them so other developers could use them.

Second, every LLM project I worked on needed the same reference material: how much does this model cost per million tokens, what's its context window, how does it compare to the alternative I'm considering. There was no single up-to-date source for this. So I built one.

Third, I wanted to build a tools site the way I think they should be built: a focused set of pages, each one useful enough that a working developer would bookmark it, rather than thin pages published at volume to chase rankings. Quality over quantity, on every page.

How the site makes money

Display advertising via Google AdSense. Ads appear after tool output and between major content sections, never above the fold or competing with the tool UI. When you turn on a browser-level ad blocker, the site keeps working — there is no anti-adblock wall.

The site also collects anonymous usage analytics via Plausible. No personal data is sold or shared. Tool inputs are never logged anywhere. See the Privacy Policy for full details.

There is no premium tier yet. If a tool ever becomes substantial enough to warrant a paid version, the free version stays functional. Paid features, if they exist, would be opt-in convenience (saved configs, team accounts, higher rate limits on the share-link feature) — never a paywall around basic functionality.

What the site does not do

  • It does not store the text you paste into the tools. Computation happens client-side.
  • It does not require accounts to use any tool. Accounts (when they exist) are optional convenience.
  • It does not show pop-ups, autoplay video, or interstitial ads.
  • It does not run on your machine when you're not on the site — no service workers, no background sync.
  • It does not detect ad blockers or nag you to disable them.
  • It does not invent "expert author" personas with fake bios.

How content gets made

Every tool page has 800-1,500 words of supporting content: what the tool does, how to use it, the underlying concept, and the questions people actually ask. Each page is written and edited by hand before publishing, with a hard rule against filler, fake credentials, and the kind of templated phrasing that pads a page without helping the reader.

The model pages and comparisons are built from structured data in a database: specs from official documentation, benchmark scores from cited primary sources. Each page leads with a unique data table, and the surrounding text reflects what is actually different about that model, so no two pages read the same.

What's on the roadmap

In rough order: more tools (target is 30+ shipped tools, then a steady 5-10 per month), expansion of the model database as new releases happen, the prompts library with user submissions, and accounts for users who want history and saved configurations across devices. The site launched in May 2026; expect the catalog to be much larger six months from now.

Contact

Email [email protected] or use the contact form. I read every message and reply to most within a few days. If you've found a bug in a tool, a wrong number on a model page, or a feature that would actually be useful, those are the messages I prioritize.