Thremark started as a personal frustration: every AI chat forgets you. You repeat yourself, copy-paste context, and start over every session. I wanted an AI that remembers — not as a feature, but as the foundation.
So I built it. Solo. From the first line of Swift to the Cloudflare backend, evaluation suite, and this website. Every decision, every pixel, every architecture call — one person, full stack.
SwiftUI, Cloudflare Workers, D1 database. Architecture designed for memory-first AI — marks, sessions, real-time sync.
Imprint voice layer (OpenAI + Gemini real-time). File generation. Image generation. Web search integration. Multi-model Smart Routing.
Siri, Widgets, Dynamic Island, Spotlight, Handoff, StandBy, Focus Filters, Share Extension, Live Activities, Liquid Glass. Full native experience.
Eval system (622 AI quality scenarios). Website + web app. TestFlight beta with real users. Subscription system (StoreKit 2 + server-side validation).
Memory is not a feature — it’s the product. Most AI apps add memory as a toggle or a settings page. In Thremark, every fact becomes a visible, editable Mark. You control what the AI knows — always.
Native, not wrapped. Thremark is pure Swift with 30 Apple integrations. Not a WebView. Not Electron. Not a React Native wrapper. Native because native is faster, lighter, and respects the platform.
Transparent by design. You see your data. You see your marks. You pick your vibe — the model is chosen for you, but you always know what’s happening. “We pick the best model” is not a black box here — it’s Smart Routing with clear logic.
Want to follow the journey?
I build in public. Updates, behind-the-scenes, and honest takes.