Real products, across the whole software stack.

Waguri is a production AI platform for Discord communities. It runs automated image (NSFW) and text moderation, holds live spoken voice conversations, learns each server's rules, handles appeals with an audit trail, and ships a full web dashboard with subscriptions and per-server billing. It is deployed in real communities and runs continuously on our own hardware, with pay-as-you-go inference as an overflow tier and circuit-breaker failover between providers.

• Vision + language models for moderation, self-hosted first with cloud failover
• Real-time voice: speech-to-text, an in-house webmail-grade IMAP mail stack, and streaming text-to-speech
• Next.js dashboard, Stripe billing, per-server configuration, appeals and audit logs

Kurono is a hybrid operating system built from the low level up in C. It ships its own desktop environment and its own language, KCL, and is designed to run native Kurono applications alongside Windows and Linux applications on the same machine. It is our most ambitious systems project and the clearest statement of how deep we are willing to go.

• Custom desktop environment
• KCL, a first-party language for native apps
• Compatibility layers for Windows and Linux software

Kaiten brings NVIDIA DLSS 4, both Super Resolution and Frame Generation, to Minecraft Java 1.21, built on a VulkanMod fork and the NVIDIA Streamline SDK. It is real graphics engineering: wiring a modern upscaling and frame-generation pipeline into a game that was never designed for it.

• DLSS 4 Super Resolution + Frame Generation
• Built on a VulkanMod fork and NVIDIA Streamline
• Targets Minecraft Java 1.21

Anime Hideout is a community platform with member profiles, moderation tooling, appeals, and a companion desktop application with its own public installer and auto-update feed. It is where much of our web and platform craft was first proven, and it still runs today.

A system-wide Windows autocorrect that recovers heavily mangled typos in real time, using a four-layer pipeline: personal memory, a keyboard and phonetic matcher, and a local GPU language model. On-device, private, and fast enough to run as you type.

Our own mail platform: a Stalwart mail server scoring a perfect deliverability rating, fronted by a from-scratch IMAP/SMTP driver we wrote for an open-source webmail client, plus a multi-tenant domains-and-addresses control panel. We run mail for our own domains rather than renting it, because owning the stack is the whole idea.