60 Days of Building
What happens when a small team ships fast across compilers, AI products, and new launches — all at the same time.
The last 60 days have been the most productive stretch in Skelpo's history. Between late December 2025 and late February 2026, we shipped hundreds of commits across multiple projects — some brand new, some years in the making. This is the story of what we built, what broke, and what we learned.
Perry: From First Commit to Cross-Platform Native Compiler
The biggest story of the last two months is Perry. On January 22nd, we made the first commit. By February 24th — just 33 days later — we had shipped version 0.2.162, a native TypeScript compiler that produces standalone executables for six platforms.
What Perry Does
Perry compiles TypeScript directly to machine code. No Node.js, no V8, no runtime. You write TypeScript, run perry build main.ts -o app, and get a 2-5 MB native binary that starts in under a millisecond.
The architecture is built on proven Rust infrastructure: SWC for parsing TypeScript, Cranelift for code generation, and a custom HIR (High-Level Intermediate Representation) layer in between that handles type analysis and optimization.
The Journey from v0.2.0 to v0.2.162
The velocity was intense. Here are the major milestones:
| Version | Milestone | |---------|-----------| | v0.2.0 | Initial compile — immediate SIGSEGV on mysql2 | | v0.2.49 | First production worker running end-to-end | | v0.2.50 | BigInt corruption fixed with proper NaN-boxing | | v0.2.87 | String operations fully working | | v0.2.99 | JSON.stringify, object literals, recursive calls optimized | | v0.2.116 | Native macOS UI support via perry/ui module | | v0.2.128 | LICM — loop-invariant code motion, mandelbrot 48ms to 27ms | | v0.2.138 | iOS support with perry-ui-ios crate | | v0.2.147 | Mark-sweep garbage collection — bounded memory for long-running programs | | v0.2.152 | Plugin system, Canvas widget, macOS UI improvements | | v0.2.157 | 12 UI/system features, web target, notifications, keychain, multi-window | | v0.2.162 | Full cross-platform UI feature parity — all 6 platforms covered |
That's 162 versions in 33 days. Many of those were multi-hour debugging sessions — fixing NaN-boxing edge cases, tracking down Cranelift verifier errors, and wrestling with platform-specific UI frameworks.
Six Native UI Backends
By late February, Perry can compile a single TypeScript codebase to native UI applications on macOS (AppKit), iOS (UIKit), iPadOS (UIKit), Android (JNI/NDK), Linux (GTK4), and Windows (Win32). There's also a web target that generates JavaScript. Each platform uses real native widgets, not webviews or Electron.
The UI framework includes 127+ functions: buttons, text fields, toggles, sliders, navigation stacks, canvas drawing, alerts, file dialogs, notifications, keychain access, and more.
Performance That Matters
Perry uses NaN-boxing for efficient union type representation, shape-cached object allocation (5-6x faster object creation), arena-based array allocation, and loop-invariant code motion. The garbage collector uses mark-sweep with conservative stack scanning.
Real benchmarks against Node.js show 2-5 MB binaries versus 80 MB+ runtime installations, sub-millisecond startup versus hundreds of milliseconds, and measurably faster execution in compute-heavy workloads.
Real-World Validation
We didn't build Perry in a vacuum. We used it to compile one of our own production workers — the MyAIRank background job processor. That worker makes API calls to OpenAI, Anthropic, and Google, queries MySQL, processes JSON, and handles async operations. It compiled and ran successfully on Perry v0.2.49, and has been our primary integration test target ever since.
MyAIRank.io: Tracking AI Visibility
In early February, we launched MyAIRank.io — a tool that answers a question every brand is starting to ask: "Do AI models know about my business?"
MyAIRank checks how often ChatGPT, Claude, and Gemini mention your website or brand. It offers quick domain checks, comprehensive GEO (Generative Engine Optimization) audits with multilingual support, and category-based visibility leaderboards.
The stack is Next.js 16 with React 19 and Tailwind CSS 4, backed by a MySQL database and a separate background worker that processes AI queries across multiple LLM providers. The worker runs either on Node.js via PM2 or as a Perry-compiled native binary.
Within the first few weeks we added Telegram notifications for signups, audit completions, and domain checks — the kind of real-time awareness that helps a new product find its footing. We also added a comprehensive test suite (83 tests covering core utilities) and enhanced the free audit report with actionable insights.
Fascinating News: AI-Powered News Aggregation
Also launched in February: fascinating.news, an AI-powered news aggregation system that collects articles from major sources, generates AI summaries using Claude Haiku, and distributes them across Bluesky and X/Twitter.
The system has several interesting components:
AI-Generated Summaries. Every article gets a Claude Haiku-generated summary with key facts. Visitors see a teaser page with the summary before clicking through to the original source. The summaries are generated both at fetch time and on-demand as a fallback.
Native Ad System. Contextual advertising built from scratch. An AI matching engine extracts keywords from articles, scores them against advertiser creatives, and then generates unique contextual ad sentences using Claude Haiku. The ads are woven naturally into the article summaries — not inserted as banners. We onboarded 9 FinanceAds partners with cookie-based ad rotation.
X Auto Quote-Tweet Monitor. The system monitors major news accounts on X, scores each tweet for relevance, and auto-generates quote tweets with AI-written TL;DR summaries. Safeguards include global and per-account daily caps, cooldown timers, and an explicit opt-in toggle per account.
Multi-Account Bluesky Posting. Five Bluesky accounts covering US and German audiences, with time-windowed posting schedules and bot-detection filtering.
The whole thing runs on Hono with string-template HTML rendering — no React, no heavy framework. Just fast, lightweight server-side rendering deployed via PM2 to our own infrastructure.
Perry Landing Page and Demo Suite
To support Perry's launch, we built a landing page at perryts.com and a benchmark demo suite. The demo includes matrix-multiply and sort benchmarks comparing Perry against Node.js, with documented results tracking compiler improvements over time.
We also published perry-starter — a hello-world template with benchmarks — and perry-pry, an interactive environment for experimenting with the compiler.
What We Learned
Compilers teach you everything. Building Perry forced us deep into NaN-boxing, garbage collection, closure semantics, Cranelift IR, and six different platform UI frameworks. Every bug was a lesson in how languages actually work.
Ship your own dogfood early. Compiling MyAIRank's worker with Perry from version 0.2.0 onward meant every real-world bug surfaced immediately. By v0.2.49, the worker ran end-to-end. That milestone validated the entire project.
AI-powered content systems are viable. Both fascinating.news and MyAIRank demonstrate that Claude Haiku can generate useful, contextual content at scale — summaries, ad copy, audit reports — when you build the right guardrails around it.
Launch fast, iterate faster. MyAIRank went from initial commit to live product with Telegram alerts and 83 tests in under a week. Fascinating.news went from zero to five Bluesky accounts, X automation, and a native ad system in two weeks. Speed compounds when you keep your stack simple.
What's Next
Perry is heading toward production readiness for server-side applications. The UI framework already covers six platforms with full feature parity. The next milestones are expanding npm package compatibility, improving compile times, and building out the developer ecosystem.
MyAIRank is growing its user base and expanding its audit capabilities. Fascinating.news is scaling its source coverage and refining its ad matching algorithms.
If you're interested in any of the technologies mentioned here — whether it's compiling TypeScript to native code or building AI-powered products — get in touch. We love working on hard problems.