whoami
Software engineer working mostly on AI agent tooling and self-hosted, own-your-stack infrastructure. I like building small, sharp tools in Rust, owning every layer of my own systems, and writing about what breaks along the way.
I write regularly at jakegoldsborough.com.
tech
AI agent tooling
I build the layer between models and real work: harnesses, orchestration, evaluation, and the plumbing that lets agents contribute to real codebases.
claux — terminal agent harness
A Rust terminal agent harness for multi-provider models: tools, permissions, sessions, cost tracking, and isolated evaluations. Built from scratch to understand agent mechanics from the inside.
llm-mux — LLM multiplexer & workflow runner
Route prompts to multiple backends and orchestrate multi-step workflows. Declarative TOML workflows, parallel/first/fallback execution, apply-and-verify steps. The reproducible "Makefile for LLMs."
replaybook — durable infrastructure-agent evaluation
Evaluate infrastructure agents against isolated, stateful incidents and verify their repairs survive a restart and a full host reboot.
skillz — a package manager for agent skills
Install, share, and manage skills for Claude Code and Codex.
Self-hosted, own-your-stack
Tools for owning every layer of your own systems: capture, index, publish, infrastructure.
raft — a knowledge graph over your notes and repos
A personal knowledge graph grown from your markdown notes and git repos. Queryable from the CLI, walkable by agents over MCP. Deterministic, read-only, and fused with git history.
webstead — federated personal site platform
A federated personal-site platform with built-in ActivityPub and threaded conversations. "Discourse for one person."
pond-nix — declarative VPS infrastructure
The NixOS config for my VPS: Gitea, GoatCounter, Caddy, and a growing set of small self-hosted services, all declarative with atomic rollbacks.
lemma — a language-course compiler
Turns structured language data (YAML) into lessons and Anki decks. Language-agnostic: every language is data, not code.
Software that doesn't have to justify itself
Some things I build because they're funny, or beautiful, or just to see if they work. Here's why.
petri — Conway's Game of Life inside n8n
An artificial-life experiment where n8n is the physics engine, the database, the display, and a resident Victorian naturalist filing field reports.
glow — give your coding agent a physical presence
Drive a VIA/QMK keyboard's RGB from the host, so a big button on my desk glows with whatever the agent is doing. No firmware flash.
schrödinger's life — a creature that dies when unobserved
A collective website whose creature is alive only while someone is watching it. When the last observer leaves, it dies permanently.
Interests
- Rust: CLIs, TUIs, systems programming, small sharp tools
- AI agent tooling: harnesses, orchestration, evaluation, MCP
- Self-hosting and declarative infrastructure (NixOS)
- The open, own-your-stack web (ActivityPub, POSSE)
- Writing about the process, failures included
More at jakegoldsborough.com · all repos.