Code stopped being my bottleneck.
Now it's ideas and feedback.
localvoxtral catches my ideas before I lose them. Users and automated checks tell me whether the result is right.
Now: vidtheque, conference talks turned into answers a coding agent can cite, down to the second.
Selected work
- StarbridgeTypeScript · Next.js · Kotlin
When a coding agent stops for a question, Starbridge puts it on your phone and in your browser. You answer with one tap and the session carries on. It works with Claude Code, Codex, Pi and opencode, and shows what’s left on each AI plan.
- localvoxtralSwift
Realtime, fully local dictation in the macOS menu bar: words appear while you’re still speaking. Built for prompting coding agents by voice, so it joins the exact Claude Code session under your cursor and polishes against what is on its screen.
- vidthequePython · Next.js · React
The talks you don’t have time to watch, turned into knowledge an agent can query mid-task. Every answer carries its receipt: the sentence, the slide, and the second it happened. Self-hosted, on hardware you already own.
- working-setPython
How many agents a given GPU configuration keeps warm, and which constraint binds first: KV cache, decode bandwidth, or prefill compute. A scenario model, and an explorer that answers with a verdict and the vLLM flags to match.
Work I sent to projects other people depend on: mlx-audio-swift, eight merged PRs that made Voxtral streaming faster than realtime and steady for three hours; llama.cpp, a merged fix that keeps Claude Code’s thinking blocks intact through the server; CodexBar, five merged PRs for Mistral plans, in two days; mlx-lm, two prompt-cache bugs, reported with repros; Node, a V8 crash on small-ICU builds, reported with a patch and a fallback package.