Control planes for machines. Tiny worlds for humans. Every system shows its work.
RUN MUTX · PLAY STARFIELD · ASK TABLEBEAM · SUMMON CKITTY · PLAY SOLAR DRIFT
I build systems that have to survive contact with reality: bounded AI agents, local-first tools, inspectable market infrastructure, and deterministic terminal experiments.
The common thread is simple. A useful system should make its state visible, preserve a trail, and let a human understand what happened after the demo ends.
Now shipping: MUTX
The control plane around the agent. MUTX gives operators one surface for identity, runs, permissions, budgets, approvals, observability, and audit history.
5 releases signed macOS app API + SDK + CLI + TUI auditable runs
Try it · Read the code · Open the docs · Download for macOS
Four orphaned prototypes, rebuilt into public proof:
A zero-network Snake roguelite with daily seeds, three run profiles, modular builds, local progression, and procedural WebAudio.
|
A seeded routing game where one closed loop, expiring deliveries, hazards, and three graph splices decide the run.
|
A local-first operator console with a visible policy gate, stoppable execution, deterministic evidence, and body-free receipts.
|
A seeded discrete-event lab for queue pressure, stage capacity, retries, and recovery—with every unit visibly synthetic.
|
The same revival pass turned two more prototypes into honest evidence labs: Barter v2 replays deterministic trade protocols with no external writes, and Ambient Relay v1 routes synthetic Discord events through explicit policy without contacting Discord or sending a message.
A zero-dependency deep-space combat roguelite with seeded runs, machine-readable state, exact-width rendering, and a tested 15-wave campaign.
|
A tiny native terminal cat in C11 and ncurses. Procedural poses, deterministic frames, resize-safe animation, sanitizers, and no runtime baggage.
|
Private-by-default Q&A for CSV and Google Sheets. It runs against local models and points every answer back to the source rows.
|
Architecture tours, demo recipes, tradeoffs, and build receipts for the systems in this workshop. The depth layer behind the demos.
|
| Principle | What it means here | Proof in the workshop |
|---|---|---|
| Bounded | Authority, budgets, and approvals are explicit. | MUTX policy and run controls |
| Local | Your data can stay on your machine. | Tablebeam and private research workbenches |
| Deterministic | Important state can be replayed and tested. | Starfield seeds, ckitty frames, fixture-driven tools |
| Verifiable | Claims have receipts, sources, or an audit trail. | MUTX events, cited table rows, release checks |
I also build local RNA review tooling, cross-border settlement and governance infrastructure, autonomous market-research agents, and multilingual client products. The repositories stay private; the architecture, boundaries, and lessons surface in the Field Manual.
Languages — Python, TypeScript, Solidity, C, SQL, Bash
Product — FastAPI, React, Next.js, PostgreSQL, Redis, Docker
Infrastructure — Terraform, Ansible, Helm, CI/CD, observability
Human languages — English, Italian, Spanish, Portuguese; basic German
If you are building agent infrastructure, local-first AI, or economically serious onchain systems, talk to me.