Products
The tools I build to work with AI.
A growing set of local-first tools, inference runtimes and agent infrastructure. Every product is built by me, runnable locally, and measured — not guessed.
Tech stack · why Rust
Rust isn't an aesthetic choice. It's the infrastructure of trust.
Every property of Rust is a piece of governance moved from “I hope” to “guaranteed.” It's the principle that guides what I build — not a dogma. The right tool for the goal, and for most of what I do, that's Rust.
Fearless concurrency
Data races are ruled out at compile time. A multi-user service with an entire class of concurrency bugs removed = reliability a company can certify.
One static binary, no runtime
No interpreter, no VM, no system dependencies. A single binary that runs behind your firewall, offline. Local-first made concrete — your data never leaves.
If it compiles, it's largely correct
The compiler moves to compile time errors that elsewhere explode at runtime. “Certify, don't trust” applied to the language itself: the compiler is the first governance gate.
A verifier for AI agents
Strong typing means an agent's code either compiles or it doesn't — no ambiguity. The compiler becomes an automatic verifier of the agent's output. Not just how I write, but how I keep an agent honest.
Beyond classic OOP: composition over inheritance. The good parts of objects (encapsulation, polymorphism via traits) combined with the functional mindset (immutability, algebraic types, pattern matching) — without the fragility of inheritance.
Dev Tools
Rust Toolchain for Local Coding Agents
Development toolchain in Rust for running coding agents locally — the foundation layer for the local-first agent workflow.
Learn more →
Agent Infrastructure
Typed Git for Coding Agents
A typed MCP server that gives agents Git as structured JSON, with two safety gates before any write — and the Git backend of the future NXM TUI.
Learn more →
Inference Runtime
LLM Inference Engine for Apple Silicon
A small, honest LLM inference engine in Rust on Apple MLX. Token-exact with mlx_lm, OpenAI-compatible, fully local.
Learn more →