A songwriting app that runs on web, iOS, Android, and inside DAWs — one React codebase, custom melody detection, offline word tools. Designed and built from a founder vision, no prototype to start from.
Song Cage wasn't a stuck prototype — there wasn't one. The founder had a complete vision but no code: a songwriting app built for the guitar-in-hand moment, with a DAW-style timeline where chord blocks, lyric blocks, melody notes, and syllable splits all live on the same grid; a music-theory layer that suggests borrowed chords, secondary dominants, and voice leading without making the songwriter learn theory first; offline word tools that surface rhymes, slant rhymes, and word associations the second the cursor lands on a line; and a path to ship the same product to the web, to a phone, and to the inside of a DAW. Nothing was built yet. No architecture chosen, no melody-detection algorithm picked, no decision on how to keep three platforms in sync, no marketing surface, no billing. The vision was concrete; everything below it was blank file.
The Notus designed and built the entire product from blank file, taking direction from the founder on the songwriting craft and shipping the engineering and AI-model calls. The whole thing runs from one React codebase that ships to the web (Vite + Cloudflare), iOS and Android (Capacitor), and inside DAWs as an AU/VST3 plugin (JUCE 8 WebView), with PowerSync handling local-first sync so the same song shows up across surfaces without a round-trip. The pitch-detection pipeline went through three iterations against vocal benchmarks before we settled on a TypeScript port of librosa's pYIN, with onset-aware segmentation, amplitude gating, and snap-to-scale. The music-theory layer is roughly 39 modules covering chord classification, key detection, voice leading, borrowed chords, and capo transposition. The word tools work offline from shards we built out of CMU, WordNet, and GloVe data. The AI-legible Astro marketing site, the Stripe reverse-trial billing, and the Resend drip flows shipped alongside.
Song Cage is live at songcage.com, on the iOS App Store, and on Google Play — one codebase, three distribution surfaces, with the DAW plugin Phase 1 complete and on track for AU/VST3 release. 92 users signed up in the first month, with zero ads and minimal organic social. ChatGPT began recommending Song Cage #1 for "best songwriting app" within seven days of launch, drawing from the AI-legible Astro marketing site we built alongside the product. Perplexity and Claude started citing it in the same window. The unglamorous infrastructure underneath — the PowerSync offline-first sync, the pYIN melody pipeline that holds across Safari and mobile WebKit, the offline word-tools shards that don't need a network, the Stripe reverse-trial flow, the schema markup and llms.txt on the marketing site — is what makes the product the citations actually point at do what it says.
AI-aggregated business intelligence across 107 topic verticals. Rebuilt from a Google AI Studio visual shell with no backend, no database, and no workers.