01
AI agents & computer use
Agents that see the screen, click and type — in a sandboxed desktop or on yours. Plugged into any model through MCP.
Research & development lab
YV17Labs — Tools for what's next
We explore new technologies and turn them into useful tools.
ls ./projects01The lab
We build tools wherever something is missing, to meet real needs. Today, mostly around artificial intelligence.
- Applied research
- Real needs
- Model-agnostic
- Used every day
02Focus
Where we're exploring.
What we're working on today. More will come.
02
Backend infrastructure
Backends where security is structural, not a matter of discipline. Small memory footprint, fast cold starts, modular by design.
03
Local AI
Models that run on your own machine — Apple Silicon or NVIDIA — behind the APIs your tools already speak, or right in the browser on your GPU.
03Projects
What we ship.
Some of our projects live on GitHub. The bigger ones get a home of their own.
Flagship
GhostDesk
- Rust
- MCP
- Docker
An MCP server that gives any AI agent eyes, hands and a full desktop. It sees the screen, clicks, types and runs apps like a human — in a disposable sandbox or on the machine in front of you.
- Linux · macOS · Windows
- Any MCP client
- Any model
TokenShooter
- WebGPU
- Transformers.js
- three.js
A small language model plays a first-person shooter, right in your browser. No server: Qwen3.5 runs on your GPU through WebGPU and, Jev-like, every move is a single token.
NestRS
- Rust
- Framework
The Rust framework for modular, scalable backends. You write the business logic; modules, dependency injection and multi-tenant security come built in.
SpecterChat
- Flutter
- MCP
- Desktop
A lightweight desktop chat client for any OpenAI-compatible API — MCP tools with their screenshots shown inline, plus image generation and editing.
Pictor
- Python
- Diffusers
An OpenAI-compatible gateway for local image models. Generate and edit images from any chat client, on Apple Silicon or NVIDIA. Ships Qwen-Image.
And more on GitHub
Forks, experiments and work in progress around llama.cpp, MLX and local inference.
github.com/YV17labs ↗Built to work together: GhostDesk runs on NestRS, and SpecterChat can drive GhostDesk over MCP and generate images through Pictor.
04Contact
Let's build something.
A question, a collaboration, or an enterprise need around one of our projects? Write to us.
hello [at] yv17labs.com