SEER
TestFlightFlagship client for Ollama Cloud — chat with frontier open models, codebases, and projects from your phone or Mac. Built-in security model and a polished, fast native UI.
TestFlightBuilder Portfolio
GTM operator who ships.
I run go-to-market by day and build the whole stack by night — native iOS apps, AI systems, autonomous agents, and the infrastructure they run on. Everything below is mine, end to end: design, code, deploy, operate. Most of it runs on a single VPS and a fleet of Cloudflare Workers.
Native SwiftUI apps, each shipped to TestFlight (several Rust- or SpriteKit-cored) through a no-Mac CI pipeline I built — App Store Connect API key + XcodeGen + a hosted macOS runner.
Flagship client for Ollama Cloud — chat with frontier open models, codebases, and projects from your phone or Mac. Built-in security model and a polished, fast native UI.
TestFlightEnd-to-end encrypted messenger (PQXDH + double ratchet, recovery phrase, at-rest encryption) with AI built in — BYOK + on-device, dedicated AI chats and @mentions inside E2EE groups. Rust crypto core compiled to an xcframework.
TestFlight21-questions, inverted: the AI reads who *you* are from your answers and paints a self-portrait back. Plus a classic word-guesser mode.
TestFlightDaily competitive word climb — bank your points or push your luck for a bigger multiplier. Deterministic engine, server-authoritative scoring, global leaderboards, Sign in with Apple.
TestFlightDaily cascade puzzle — Drop7's chain reactions fused with push-your-luck greed. All-integer deterministic engine projected through a SpriteKit scene with escalating juice.
TestFlightOne-finger maze runner — drag from start to goal in a single unbroken motion; hit a wall and slide back along your own trail. Pure-Swift, Linux-testable engine.
TestFlightAn AI-first persuasion game: talk a fictional AI gatekeeper into bending its own rules. A study in prompt architecture and resist-vs-yield asymmetry, dressed as a game.
TestFlightAI debate arena — pit models against each other on any motion and judge the round. Native app plus a full web mirror sharing one backend.
Open web appA community companion concept — my one React Native / Expo build on a Cloudflare Worker + D1 backend, exploring a different app stack end to end.
—Personal client that routes Ollama Cloud through a Tailscale-secured VPS gateway — a thin app over self-hosted AI infrastructure.
—Production web apps, each a Cloudflare Worker (+ D1 / Stripe where needed) on my own domain.
BYOK chat over the top open models with a SEER-styled UI. The hub of the whole stack.
VisitFree, gated mirror of manticthink — server-key chat for the top models, with premium models + codebase chat behind BYOK or a Stripe subscription.
VisitFull AI-debate arena in the browser — free daily cap plus a Stripe Pro tier, sharing the native app's backend.
VisitCustom coding agents, fine-grained model personalities, and an autonomous agent that lives on its own — the experimental edge of the stack.
kimi-code and minimax — bespoke coding agents wired into the Codex CLI on top of open models, each with its own system prompt, tools, and temperament.
—An unguided coding agent jailed in its own sandbox, woken on a timer to do whatever it decides — its persistent process is watchable live on the public web.
Watch liveHand-authored Ollama models — Herald, Cartographer, Heretic, minimax-codex — each a distinct system persona for a specific job.
—An ongoing bench of agent experiments — Babel, Ecnyss, MiniMadMax, MiniMaxine — where new ideas get prototyped before they graduate into products.
—Public, sandboxed Telegram services — each one a real, always-on product behind a chat handle.
A public bot-builder: describe a bot in chat and it provisions, codes, and runs a sandboxed Telegram bot for you. Claude-backed.
@Build_Your_AI_BotBuilds and deploys real websites by chat — describe a site, it codes and ships it to a live URL.
@AI_Web_Dev_BotKimi, MiniMax and Babel — Telegram front-ends that bridge straight into CLI coding agents, so I can ship code from my phone.
—A public chat bot routing to DeepSeek over Ollama Cloud — open-model chat for anyone.
—The plumbing under everything — a single VPS running dozens of systemd services, Cloudflare tunnels, and remote-control planes I designed and operate.
A cross-device session-jumping mesh — hop a live working session between machines. Ships with its own threat model and enhancement design docs.
—A remote Model Context Protocol server that gives Claude (web/mobile) hands-on access to the whole VPS, behind GitHub-OAuth allow-listing and fail-closed by default.
—Self-hosted 'give Claude hands on your own machine' — a custom connector you deploy on your own box and hold the keys to. Open source.
GitHubWireGuard + IKEv2 VPNs, Cloudflare tunnels, Unbound DNS, fail2ban — the personal cloud the rest of this runs on.
—Design docs and threat models written alongside the builds — the thinking, not just the shipping. (Summaries; source lives in the project repos.)
A scheme for cryptographically verifiable contact introductions in an E2EE messenger — making 'who vouched for whom' auditable without a central authority.
—The adversary model behind Logos, plus a blueprint for folding AI into an E2EE app without breaking its guarantees (BYOK, on-device, sealed sessions).
—What it means — and what it costs — to let a live session jump between devices safely.
—How to make an AI gatekeeper that resists cheap tricks but yields to genuine craft — the asymmetry that makes the game (and the safety lesson) work.
—