# Protechtus LLC, the engineering studio

> Local-first AI infrastructure that checks itself as it ages.

Last verified: 2026-11-09. Markdown version of https://protechtus.com/work.html.

Protechtus is a small engineering studio in Draper, Utah. It builds AI infrastructure that checks itself as it ages, Sextant first, and the tools around it: libraries extracted from Sextant, a suite of terminal apps, and a research library of 4,000+ verified reports that backs every guide published here. Sextant and the extracted libraries ship with tests. Every claim on this site is checked against a source or labeled.

## Who builds it

Jay Pitts. Engineer in Draper, Utah, focused on security and AI orchestration. Grew up fixing machines in the Salt Lake Valley; builds local-first AI infrastructure where correctness and resilience are designed in from the start.

| | |
|---|---|
| Focus | AI orchestration, agent memory, and security-first local infrastructure |
| Security | Weekly automated audits and encrypted, recovery-tested backup on my own infrastructure; on-device data handling |
| Studying | CompTIA Security+ and Network+, in progress |
| Certified | SSCA |
| Background | Computer repair and malware remediation since 2012; small-business IT |
| Entity | Protechtus LLC, which owns Sextant and its tooling |
| Standards | Build it once and build it right. Vendor-neutral. Local-first. Resilient by default. |

## Selected work

A few of the things, with more on the way. Everything here was designed and built end to end under Protechtus LLC. Sextant is being released now. The research library runs alongside it and backs the guides. The terminal suite and the extracted libraries follow as each is ready.

### Sextant (platform)
A local-first agent control plane. Keeps an LLM's long-term memory correct over time through epistemic decay and scheduled re-verification, and orchestrates autonomous work through a task queue, a cron scheduler, quality gates, and model-escalation routing, with local models by default and cloud models optional. Tests: 4,000+ core. Runtime: local, free. Status: v0.1.0, in daily use.

### Research library (pipeline)
A research corpus that re-verifies itself. Four thousand-plus research reports and the pipeline that keeps them honest: nearly every report carries a current verification review, is re-checked on a staleness schedule, and passes a citation guard that rejects fabricated sources before anything is promoted. Reports: 4,000+. Verified: nearly all, currently. Status: private, in daily use.

### Extracted libraries (private, may open; each v0.1.0, tested)
- **inspiration-engine**: semantic-entropy self-consistency plus a verify-before-promote gate. Turns raw model output into vetted ideas instead of trusting the first draft.
- **epistemic-decay**: the confidence-decay model and re-verification behavior as a reusable library. A fact's confidence erodes on a half-life curve and is re-checked on a schedule.
- **rpo-guard**: recovery-point-objective miss detection plus a disk-space guard. Born from a real backup-loss event. Catches silent backup failure before it costs data.

A suite of terminal apps and more internal tooling follow, named here as each is ready to be released.

## Engineering values

- Build it once, build it right. Invest in solid foundations over quick patches.
- Version before edit, archive never delete. File state is preserved before any risky change; stale material is archived with provenance, not removed.
- Durable artifact per session. Every working session produces files, updated memory, and a log.
- Vendor-neutral by default. Multiple models in parallel, no single-vendor lock-in, every artifact usable by any LLM.
- Local-first and cost-aware. Prefer local inference and free execution; spend stronger models only where they earn it.
- Resilience as a first-class concern. Assume things can vanish. Back up, document, make work reproducible.

## Get in touch

hello@protechtus.com. Protechtus LLC, Draper, Utah.