Kansas City, Missouri — Builder & Operator
I'm Christopher Moore, a builder in Kansas City. I direct a team of AI agents, build the infrastructure they run on — gateways, relays, operator tooling — and make the dashboards and evaluations that keep all of it honest. The discipline is simple: define the outcome, measure it, ship the thing people can actually use.
Selected work
Filter by what you're here for. Everything below is built, running, and used — not a mockup.
A 15-agent team with a supervisor, specialist workers, and a security reviewer — restructured in Oct 2026 around one rule: real deliverables first. Single owner per task, end to end; gates only where money, credentials, or publishing are on the line.
A local OpenAI-compatible gateway that gives every agent in the roster the same live context — identity, memory, people, and goals — injected under each agent's own role prompt, then forwards transparently to the model. One brain, many jobs.
Two ways to work with machines I own: Muse Relay (the cursor-relay) — a relay I start myself that lets an assistant see my screen and drive mouse & keyboard at user level — and a direct OpenClaw operator client for directing agents in WSL: health, sessions, and dispatch, verified end to end.
KPI-style boards, the way a business major wants them: headline metrics first, pace against the period, categories for visibility — never per-category micromanagement. One board watches every scheduled job; another tracks plan usage against its weekly limit and reset.
A Strava-like personal-best board for 11 distances, auto-computed from logged runs, with a live delta against your PB “ghost” and OpenStreetMap route maps with GPX import / export. Starts empty — your data, your board.
Showing all projects
Reference
Written up as reference pieces, not scoreboards: what was tested, how, and what the result actually changed. If an agent or a tool can't do the real task, the write-up says so.
A relay that can see a screen and send mouse and keyboard input sounds like total control. This evaluation pinned down what it really is, capability by capability — because "an assistant can drive my PC" is exactly the kind of claim that should survive a threat model, not a vibe.
When Meta shipped native computer use in Muse for Mac — control inside real apps, per-app opt-in permissions, approval before sensitive actions — I stacked it against my own screenshot-and-click relay instead of assuming the home-built tool was fine.
Benchmarks for agents shouldn't be trivia questions — they should be the actual work. The Cadre roster gets evaluated on live production tasks with checkpoint reviews, including adversarial capability tests designed to find where an agent bluffs, stalls, or over-complies.
The most useful evaluation result I've gotten wasn't a score — it was watching an agent (this one) dispute a quoted email, re-check it six times, and produce confident letter-by-letter "proof" of a reading that was simply wrong. An independent pixel-level read by a different model settled it in one pass. A checker re-running its own method mostly re-runs its own belief.
Approach
Manufacturing teaches you that a process you can't measure is a story you're telling yourself. I run software and agent teams the same way.
Milestones are user-visible and real-world-visible. Infrastructure earns its place by unblocking a deliverable — never by being impressive on a board nobody uses.
Headline number, pace against the period, categories for visibility. When usage metering lied because it read the wrong table, the fix wasn't a nicer chart — it was finding where the real data actually lived and verifying it live.
If a fact is disputed, checking it the same way again just reproduces the same belief. Get independent evidence — a second method, raw bytes, a live test. Applies to code, data, and anything with a spec.
Capabilities
Multi-agent orchestration, gateway & context design, scheduled automation, and the operating rules that keep a roster of agents shipping instead of ceremony-ing.
Small, sharp tools that get used daily: KPI boards, personal trackers, map-based apps, and CLI tooling — macro-free and web-friendly where the workplace demands it.
A business major's instinct for how work actually flows: budgets monitored like KPIs, and process fixes grounded in where the data actually lives — not where the chart assumes it does.
Benchmarks and evaluations for agents and tools: capability tests, security reviews, and head-to-head comparisons run against real tasks — with verdicts written up as reference, not marketing.
About
I'm Christopher Moore, a builder based in Kansas City. My rule for everything on this page: a result is real when you can measure it — parts are in spec or they aren't, a build runs or it doesn't.
I run a small, strange operation out of Kansas City: a team of AI agents I direct like a (sometimes chaotic) crew, the infrastructure underneath them, and a bench of tools and dashboards built because I wanted them to exist and work exactly my way.
The through-line is a business major's instinct: watch the numbers, keep the categories visible, and never let process outrank the product.
The agent roster, its infrastructure, and the tool bench — all in active production, all measured against real use.
Restructured a 15-agent system around deliverables-first work: single owners, standing preflight, gates only at real risk.
A direct control plane for machines I own: a user-level desktop relay plus a paired OpenClaw operator client, verified end to end.
Capability tests and security evaluations of the agents and tools I rely on, written up as reference pieces.
Contact
The fastest way to see what I actually do is to look: browse the code, read the evaluations, then reach out on GitHub. I'm especially interested in AI operations, automation that survives contact with real work, and small tools done properly.
MOREFOLIO.COM — REGISTERED OCT 13, 2026 · THIS SITE IS ITS HOME