DevKits
34 releases · 163+ tools shipped

What's new in DevKits

Every release, in the open. New tools, improvements, and fixes — with links to the exact pages we shipped. If you spot something missing or broken, tell us on the message board.

  1. Latest

    New: an interactive 3D guide to how LLMs work — in English and 中文

    A brand-new learning chapter that visualizes a working GPT-style transformer in 3D — watch a real (tiny) model run inference token by token, layer by layer. It's fully self-hosted (no external downloads, no tracking) and ships in both English and Chinese, with a stage-by-stage written walkthrough from tokenization all the way to sampling.

    How LLMs work — interactive visualization

  2. Getting more tools into Google's index

    A focused pass to unstick the tool pages that Google had crawled but not yet indexed. The sitemap now sends honest, spread-out freshness dates instead of dozens of pages sharing one identical timestamp, ranks actively-maintained tools above long-tail ones, and every tool page now funnels internal links toward the pages that need them most — the strongest signal that a page is worth indexing.

    Indexing & discoverability

    • Improved
      Sitemap freshness dates for tools without an explicit update date are now spread deterministically across a real date window instead of all sharing one day — removing the 'batch-updated' pattern Google reads as fake freshness and delays indexing over.
    • Improved
      Sitemap now ranks priority and change-frequency by whether a tool is actively maintained, so Google's limited crawl budget flows to the highest-value pages first instead of treating all 200+ tools as equal.
    • Improved
      Related-tools sections now steer internal links toward long-tail pages and rotate targets per page — spreading link equity across the whole catalog rather than repeatedly pointing at the same few flagship tools.

  3. Crawl-budget cleanup — sharper indexing signals

    Behind-the-scenes SEO housekeeping so Google spends its crawl budget on the pages that actually help you. Dynamic Open Graph images are no longer mistaken for standalone pages, and the thinnest programmatic reference pages (accented Latin-1 letters and unused control codes) now bow out of the index cleanly — sending consistent signals across sitemap and page metadata.

    Indexing & discoverability

  4. AI engineering toolkit — token, context, cost, JSON & schema

    A new cluster of tools for people building with LLMs. Check whether a prompt fits a model's context window before you hit 'context length exceeded', break a prompt down to see which section eats your token budget, pull clean JSON out of any ChatGPT or Claude reply, and generate OpenAI / Anthropic / Gemini function-calling definitions from one sample — all client-side. Plus two new reference pages: a searchable pricing table across 100+ models and a fix-it guide for the API errors you actually hit.

    12

    New AI tools

    4

    New reference pages

    New AI / LLM tools

    New reference pages

  5. Sharper search titles + tighter index focus

    Every tool page now shows a search title tuned for the long-tail queries a new site can actually win — leading with what makes the tool different (no upload, offline, client-side, instant) instead of a brand suffix nobody searches for. At the same time, the emptiest ASCII pages (C0 control codes and the unassigned C1 block) are now kept out of the index so Google spends its crawl budget on the pages that matter.

    Long-tail search titles

    • Improved
      Tool pages without a hand-written search title now generate one automatically that leads with the tool's edge — 'No Upload', 'Offline', '100% Client-Side', 'Instant' — so the page competes for specific, winnable queries instead of head terms owned by decade-old sites.

    Tighter index focus

    Sharper titles & FAQs on core tools

    • Improved
      The most-used tools — JSON Formatter, Base64, JWT Decoder, Hash Generator, UUID Generator, Timestamp Converter, and Regex Tester — got search titles and keywords tuned to real long-tail queries (format large JSON, file to Base64, decode a JWT without the secret, epoch in milliseconds, bulk UUIDs) plus extra FAQs that answer those questions directly.

  6. Deeper ASCII reference pages + sharper HTTP header search snippets

    Every one of the 256 ASCII / Latin-1 code-point pages now carries unique, in-depth content — a 'Common uses & context' write-up, category-aware programming notes, and three-plus code-specific FAQs — instead of a shared template. The Expires and Content-Disposition header pages also got intent-matched titles, aliases, and search snippets so the queries they already rank for finally turn impressions into clicks.

    Richer ASCII table pages

    HTTP header search snippets

    MIME type & regex reference FAQs

  7. Fix: JSON Formatter fullscreen no longer renders a black screen

    Entering true fullscreen on the JSON Formatter used to leave you staring at a black void — the editor panels collapsed to zero height because their fixed 56rem height couldn't resolve inside the fullscreen flex/grid layout, exposing the browser's default black backdrop. Fullscreen now stretches both the input and output editors edge-to-edge and forces the themed background over any user-agent fullscreen styling.

    Faster homepage

    • Perf
      The homepage now server-renders only the top 6 tools per category (small categories stay full) instead of all ~197 tools at once. This cut the raw HTML from ~821 KB to ~567 KB and the inline React payload from ~639 KB to ~340 KB — the first paint on a cold visit lands noticeably sooner. Every hidden tool is still one click away via each category's "View all" hub.

    Fullscreen editing

    Indexing & internal links

  8. GSC sweep — surface pages Google is already showing but nobody clicks

    Cross-referenced the last7 days of Search Console impressions and found 14 queries getting served but no clicks. Half the fixes are pure metadata (titles, keywords, aliases) — no new pages, but every affected URL now speaks the same language as the actual queries. Also shipped a client-side ASCII lookup so `0x41`, `×`, and `U+00D7` all resolve to the right character page.

    GSC-guided metadata rewrites

    ASCII lookup

    HTTP header multi-language & alias support

    Sitemap freshness signal

    • Improved
      CONTENT_LAST_UPDATED bumped from 2026-07-20 to 2026-08-03 — the whole PSEO corpus (500+ pages: MIME, HTTP status, HTTP headers, ASCII, learn concepts) now reports a fresh <lastmod> to Google, prompting a re-crawl that will pick up all the alias / keyword changes.

    HTTP status code aliases

    Internal linking + H1 tune

  9. JSON Formatter — smart typing (auto-pair, Tab indent, Enter split)

    Typing JSON by hand no longer feels like a plain textarea. Brackets and quotes auto-close, Tab actually indents, Enter inside `{}` / `[]` opens a nested block with the right indent, and Backspace on empty pairs removes both sides at once. IME (Chinese/Japanese) input is untouched, and native Undo/Redo still works because every change goes through `execCommand('insertText')`. Also fixed the fullscreen mode so the first activation no longer flashes a black backdrop.

    Smart input

    Fullscreen fix

    Auto-indent while typing

  10. Fix Number to Words — was truncating at every 'thousand' boundary

    The toWords() recursive function for numbers ≥ 1000 had a logic bug that dropped every tail digit: 1234 used to render as 'one thousand' instead of 'one thousand two hundred thirty-four'. Rewrote the algorithm to split into 3-digit groups and combine with the right scale.

    Bug fix

  11. SEO speed sprint — inline CSS + deferred analytics + modern browser target

    Shaved ~1.5s off LCP/FCP: CSS is now inlined (no render-blocking request), all three analytics are deferred to requestIdleCallback, and SWC targets Chrome 100+ / Safari 16+ / Firefox 100+ for lighter JS output.

    Performance

    • Perf
      Experimental inlineCss enabled: all Tailwind CSS (80 KB raw) is now inlined in the HTML <head>. No more external render-blocking CSS request. Brotli-compressed HTML actually shrank from 59 KB to 49 KB because CSS and HTML share a compression dictionary.
    • Perf
      Vercel Analytics, SpeedInsights, and Ahrefs are now lazy-loaded via requestIdleCallback. They no longer compete for main-thread time during the critical rendering path, directly reducing TBT.
    • Perf
      Added browserslist targeting Chrome 100+ / Safari 16+ / Firefox 100+ / Edge 100+, which tells SWC to skip unnecessary legacy transforms and produce smaller JS output.

  12. Homepage performance — lighter client bundle, no more prefetch waterfall

    Cut the homepage first-load JS by ~230 KB and eliminated the RSC prefetch waterfall that was making every visit feel like it was 'still loading'. The tool grid, recently-released strip, and search results now share a tiny search-only index instead of pulling in the entire 332 KB tool registry, and every internal link across the site now opts out of auto-prefetch by default.

    Homepage performance

    • Perf
      Client bundle no longer imports the full tool registry — Recent Tools, homepage search results, and Recently Released now share a lightweight search-only index (~4 KB) instead of 332 KB of tool metadata.
    • Perf
      Every internal <Link> across the site now defaults to lazy prefetch via a new PrefetchLink wrapper: no more auto-prefetch on entering the viewport (removes the 100+ ?_rsc= fetches that were firing every time you loaded a page), but on hover / focus / touchstart we still fire router.prefetch() so click-to-navigate feels instant.
    • Perf
      Split Recently Released into a server component (renders all 12 links in HTML for SEO/crawlers) plus a tiny client toggle — the interactive shell no longer pulls tool data into the browser.

  13. SEO CTR sprint — GSC-optimized titles, doubling internal links, + new pillar concept

    Analysed GSC keyword data (60+ queries, 0% CTR, avg position 20–90) and shipped the highest-ROI fixes. Seven tools got GSC-optimized <title> tags; 10 PSEO pages had related-link density doubled from 3 to 8 links; 3 hub pages gained ItemList structured data; and a new /learn/http-redirects pillar concept page was published.

    7

    Tools with CTR-optimized titles

    10

    Pages with expanded related links

    1

    New pillar concept

    3

    Hub pages with ItemList schema

    Search-optimized titles for top GSC queries

    Build fix — MIME type redirects corrected

    • Fixed
      /mime-types/json redirect destination corrected from /mime-types/application/json to /mime-types/application-json (the actual MIME slug).
    • Fixed
      /mime-types/svg redirect destination corrected from /mime-types/image/svg+xml to /mime-types/image-svg-xml — the old path contained a literal '+' that Next.js was parsing as a regex quantifier, crashing the production build.

    Internal linking density doubled — related headers, MIME types, and status codes

    New pillar concept page + pillar→PSEO internal links

    Site-wide content expansion — 5 comparisons, 3 Best-of pages, self-hosted search, HTML sitemap + privacy architecture page

    Discovery features — MIME quick reference, Try Next chain, Tool of the Week

  14. Cross-linking pass — hub pages funnel into tools, homepage surfaces recent releases

    A traffic-plumbing pass across the site. Every MIME-type, regex-cookbook, and HTTP-status-code detail page now suggests the right hands-on tools; the homepage gets a server-rendered "Recently released" strip; seven high-traffic security & AI tools got long-form explainer content, and a dozen top pages got sharper titles with clearer free/no-signup/local signals.

    700+

    Pages with new related-tools blocks

    +7

    Tools with deep long-form content

    12

    Sharpened titles / descriptions

    New long-form guides on the top security & AI tools

    Hub → tools cross-linking

    Homepage & discoverability

    GSC-informed SEO fixes — capturing 0-click queries

  15. One-click IndexNow submit + daily auto-cron

    Every URL from the sitemap is now pushed to Bing / Yandex / Naver / Seznam / Yep automatically once a day, and can also be triggered manually from the admin panel after shipping new tools.

    Automation

    • New
      Vercel Cron runs /api/cron/indexnow every day at 04:00 UTC — fetches the live sitemap, submits all URLs in batches of 500 to IndexNow, and logs a summary to the Vercel cron dashboard.
    • Privacy
      Cron endpoint gated by Authorization: Bearer $CRON_SECRET (same pattern as refresh-pricing) — reusing the existing secret, no extra config needed.

    Admin experience

    • New
      /admin now serves a landing page with cards for Comment moderation and IndexNow submit.
    • New
      /admin/indexnow — preview all sitemap URLs (host, shard breakdown, sample list), verify the IndexNow key file is reachable, then submit all URLs in batches of 500 with per-batch status feedback. Now also shows the daily auto-cron badge.
    • Improved
      Login now lands on /admin (was /admin/messages); every admin sub-page has a back-to-admin link.

    Under the hood

    • New
      Shared src/lib/seo/indexnow.ts powers the CLI script (pnpm indexnow), the admin API, and the cron endpoint — sitemap-index parsing, key verification, batched POSTs.
    • Privacy
      Admin API rejects unauthenticated calls; key file is loaded from INDEXNOW_KEY env var first, then falls back to the committed public/<hex>.txt.

  16. SEO backbone upgrade — schema, RSS, breadcrumbs, branded 404

    Six coordinated SEO improvements that make DevKits easier for Google to understand, rank, and surface. Nothing changes for users on the surface, but search engines now see a much richer picture of the site.

    Structured data

    • Improved
      Homepage now emits a full Organization JSON-LD (logo, description, sameAs) so Google can build a knowledge-panel-ready entity for the site.
    • Improved
      Every HTTP status code detail page (60+ pages) gets a BreadcrumbList schema so SERP listings show 'Home › HTTP Status Codes › 404 Not Found' instead of a raw URL.
    • Improved
      Every 'Days Until X' page adds a BreadcrumbList alongside its Event schema.

    Discoverability

    UX + edge cases

    • New
      Branded 404 page with quick links to 24 popular tools and all category hubs — plus robots noindex so typo traffic doesn't pollute the index.
    • Improved
      Homepage tools grid now has a stable #tools anchor for deep-linking from 404s, the header, and future CTAs.
    • Perf
      Sitemap and RSS routes marked force-static — pre-rendered at build time, zero runtime cost.

  17. Community Board improvements — optional tool, auto nickname

    The comment experience got two quality-of-life upgrades and a discoverability boost: choosing a tool is now optional, nicknames auto-suggest based on your rough location, and every tool page shows a Comments icon at the top.

    Comment form

    Discoverability

    • New
      Every tool page shows a 'Comments (N)' icon button in the page hero that scrolls to and auto-expands the discussion — comments were hidden at the bottom before.
    • Perf
      New lightweight /api/messages?count=1 endpoint (60s cache) powers the count badge without loading the full thread.

  18. Security & data suite — 13 new tools + Diff URL sharing

    A full production-grade crypto workflow (key generation → sign/encrypt → verify → JWKS publishing), plus CSS layout visualizers, a fake-data generator, and shareable diff links. All 100% client-side.

    13

    New tools

    20+

    New crypto algorithms exposed

    Asymmetric crypto suite

    JWT overhaul

    Data & IDs

    CSS & diff

  19. Unified page hero and consistent widths across the site

    Every non-tool page (Changelog, Community, Comparisons, About, Category hubs, cheatsheets, references) now shares the same signature gradient hero and a consistent max-width, matching the home page and the tool pages.

    Design consistency

    • Improved
      Introduced a shared PageHero component with the brand radial-gradient — same look as the home hero across Changelog, Community Board, /compare, /visualize, category hubs, cheatsheets, references, About/Privacy/Terms/Contact.
    • Improved
      Aligned page container width to max-w-6xl on every page — the horizontal edges now line up cleanly with the header and footer instead of jumping between 3xl / 4xl / 5xl / 6xl.
    • Improved
      Long-form pages (About / Privacy / Terms / Contact) keep a comfortable 3xl reading column inside the wider frame, so line lengths stay readable.

  20. Image Toolkit — 19 privacy-first image tools

    A complete image utility suite, from EXIF forensics to format conversion, resizing, cropping, favicon generation, SVG optimization, and color extraction. Every operation runs locally.

    19

    New image tools

    None

    Uploads required

    Privacy essentials

    Format conversion

    Transforms

    Developer specialties

    Under the hood

    • Privacy
      Every image operation runs 100% locally — files never leave your browser
    • Improved
      Shared image lib (Canvas helpers, format matrix, dropzone) keeps every tool page < 6 KB

  21. PDF Toolkit — 8 privacy-first PDF tools

    A full suite of PDF utilities that run 100% in your browser. No upload, no watermark, no sign-up. Verified on real-world CJK-heavy PDFs.

    8

    New PDF tools

    cmaps + fonts

    Static assets shipped

    New PDF tools

    Quality & reliability

    • Fixed
      Fixed pdf.js worker loading so all PDF tools work out of the box
    • Improved
      Text extractor now handles Chinese/Japanese/Korean PDFs correctly (via cmaps + standard fonts)
    • Improved
      Reading order rebuilt from geometry (x/y transform) instead of pdf.js hasEOL, so paragraphs no longer get split at random
    • Privacy
      Every PDF operation is 100% local — the file never leaves your device

  22. Random, Markdown & speech utilities

    Rounded out the generators category with cryptographic-strength random tools, a Web Speech API text-to-speech, plus Markdown authoring helpers.

    New generators

    Markdown

  27. JSON expansion — 15 new tools + visualizer

    Full JSON stack: repair, sort, flatten, JSONPath tester, visualizer, plus JSON → Python / PHP / Java / C# / Rust converters.

    New tools

    Visualizers & comparisons

  29. Editor upgrade + community board

    Shared editor with line numbers, error jumping, and in-editor search — used by JSON/YAML/XML/SQL. Plus per-tool comments and a global message board.

    Editor

    • Improved
      Line numbers + click-to-jump on errors
    • Improved
      Ctrl/⌘-F search inside the editor without leaving the page
    • New
      Foldable JSON tree view with copy-path support

    Community

  31. Bookmarks & unified layout

    One-click bookmarks on every tool page, plus a homepage relaunch that keeps tools visible above the fold.

    Bookmarks

    • New
      Star any tool from its page or the header — synced across the site via localStorage

    UI polish

    • Improved
      Unified 6xl container width from Header to Footer
    • Improved
      Task-focused 'workflows' block on the homepage replaces filler content
    • Fixed
      Automatic reload on ChunkLoadError to survive deployment cache races

  32. SEO backbone — Cookbook, Headers, Comparisons, ASCII, Visualizers

    Hundreds of reference pages designed for long-tail search: HTTP headers, regex patterns, MIME types, ASCII table, and comparison guides.

    300+

    Reference pages added

    References

    Visualizers & comparisons

    SEO structure

    • New
      9 category hub pages targeting mid-volume queries
    • New
      Deep tool content (about, howTo, use cases, code examples, tips) rendered on every tool page
    • New
      Smart 'Related tools' ranked by category + keyword overlap
    • Improved
      Dynamic OG images generated per tool and per HTTP status code

  33. JWT, HMAC, HTTP status codes & more

    Security tooling plus the first big PSEO project: 51 individual pages for HTTP status codes.

    New tools

    PSEO

  34. Launch — 25 privacy-first developer tools

    The initial DevKits release. Every tool runs 100% in the browser. No accounts. No uploads. No tracking of your inputs.

    25

    Launch tools

    Nowhere

    Where your data goes

    Foundations

    Site infra

    • Privacy
      Every operation runs client-side — no data leaves your browser
    • Perf
      HSTS, permissions-policy, aggressive caching for static assets
    • Improved
      Google/Bing site verification, canonical URLs, sitemap, robots.txt
Our commitment

We ship in the open. Every user-visible change goes into this changelog — new tools, behavior tweaks, security posture, and bug fixes. We don't quietly harvest data, we don't add trackers behind your back, and we don't upload your files. If a release ever changes any of that, it will be right here at the top.