Work · Co-Founder, Primary Builder · 2025 to Present

Phantomline

I wrote most of Phantomline, a local-first AI video studio for faceless YouTube creators. Co-founded with Cesar Martinez (UI work) and Zac Dillard. The whole stack runs on the user's own machine. No paid APIs, no usage caps, license keys validate offline.

What Phantomline is

Phantomline is a local-first AI video studio aimed at faceless YouTube creators. The typical faceless workflow today requires a script generator, a voiceover tool, a music library, a stock visual API, a video editor, a scheduler, and a thumbnail tool. That stack is seven monthly subscriptions before a single video ships. Phantomline replaces it with one install that runs entirely on the user's hardware.

The pitch in one line: software you paid for should keep working when the company's website goes offline. License keys validate offline. The model weights live on your machine. The only thing on our servers is the marketing site.

What I built

I wrote most of the Phantomline codebase. The Flask backend, the routing architecture, the license verification system, the schema markup across the marketing site, the blog publication pipeline, the SEO infrastructure (sitemap, robots, llms.txt, IndexNow integration), the alternatives comparison pages, and the topic cluster content all came from me.

Cesar Martinez handles UI work. Zac Dillard is the third co-founder. The three of us are also the team behind ProLocalBuilder.

AI vibe coding is how I shipped most of it. Five years of LLM workflow experience compounds when you are building a product like this: Claude and Cursor running constantly, custom prompt libraries for routine work, drafts AI-assisted and human-reviewed against documented rules.

How Phantomline works

One Flask codebase serves three surfaces: the hosted PWA at phantomline.xyz, a full desktop install, and an Android APK (Capacitor wrapper). The heavy AI inference runs either server-side on the desktop install or entirely in-browser on the hosted path.

  • Desktop mode: Ollama (Llama 3.1) for scripts. Kokoro TTS for narration. MusicGen for ambient backing tracks. ffmpeg for MP4 assembly. Optional Forge/AUTOMATIC1111 for AI-generated scene art.
  • Browser mode: WebLLM (Llama 3.2 1B in WebGPU) for scripts. Web Speech API for narration. Web Audio for music. ffmpeg.wasm for MP4 assembly.
  • License system: Offline-verifiable HMAC-SHA256 keys. No phone-home, no validation server, no subscription enforcement.
  • Deployment: Render auto-deploys from main. Mobile via Capacitor wrapping the live site.

Built with Cesar and Zac

Phantomline is co-founded by three people. Each brings a different focus:

  • Kyle McCool (me), based in Appleton, Wisconsin. Primary builder. SEO, marketing, content, and most of the codebase.
  • Cesar Martinez, based in San Jose, CA. UI work.
  • Zac Dillard, based in Las Vegas, NV.

The same three of us also built ProLocalBuilder.

The design call that shaped the product

Cloud-only AI video tools break at volume. A creator shipping 30 videos a month at 4 cents per render across the script, voice, music, and caption pipeline pays roughly $36 in pure API cost per month on top of subscriptions. Scale that to multi-channel work and the meter runs $180+ monthly before anyone watches a video.

Local AI inverts the math. Once the models are installed, every render is free. The hardware the creator already owns does the work. That was the design call that shaped the entire product, and it informs every decision I make about which features ship.

Other projects