I build local-first AI developer tools, open-source streaming projects, desktop experiments, and game/mod utilities — shipped under Kuno Labs and here.
Current focus: YukiStreams, CodePrism, and Nimbus — tools that keep AI-assisted work grounded, make media and game streaming reliable for real users, and stay useful when the network is off.
|
Stremio addon for anime, Asian drama, movies, and series — with optional Real-Debrid and TorBox support. The most-used thing I ship: a public instance, real users, real uptime expectations. |
Local-first context saving and token optimization for AI coding agents: map first, slice next, retrieve exactly. Turns a repository into an inspectable graph before an assistant reads the whole tree — exports Markdown, JSON, DOT, SQLite, and a static browser viewer. |
|
Community-maintained game streaming host fork in the Vibepollo → Apollo → Sunshine lineage. Virtual display reliability, Windows capture behavior, frame pacing, Android TV compatibility, and honest release hygiene — for people who actually stream from a gaming PC. |
Companion client track in the Artemis/Moonlight Android lineage. Android TV polish, controller ergonomics, reliable pairing, and clear compatibility with Nimbus, Artemis, and Moonlight — once the Nimbus host foundation is stable. |
# CodePrism in three lines
codeprism prime "what I am changing" --changed
codeprism query "where does this behavior live?"
codeprism visualize- 🛠️ baro-hud-vitals — Barotrauma mod: crew + player health and nav HUD.
- 🎮 Godot experiments and game prototypes — cozy games, MCP-driven agent tooling for the editor.
- 🖥️ Desktop tools and local AI stacks that are practical for non-technical users.
- Local-first AI tooling that works before it talks to the network.
- Codebase maps, slices, and replay artifacts that make agent work inspectable.
- Streaming projects with honest release notes and reproducible builds.
- Desktop tools that are practical for non-technical users.
- Game and mod utilities that favor readability, stability, and fast iteration.
Map first, then read.
Keep useful artifacts inspectable.
Prefer deterministic parsing before heavy magic.
Measure before making performance or token-saving claims.
Make public maintenance boring, transparent, and repeatable.
Build tools that stay useful when the network is off.
