Index
/epiclavis:indexAsk what a change will break, whether something already exists and what you noted last time.
epiClavis is a design-control engine for software written with AI. It tracks how the code depends on itself and checks each edit before it lands. It also keeps decisions, requirements and evidence as a live record rather than documents nobody reopens.
I use it on my own projects, including epiClavis itself. The decisions behind this page, the requirements for epiEspañol and the override records for epiClavis's own frozen specifications are all kept in it. Daily use on real projects is how I test whether it holds up.
01 · One request, start to finish
A request goes through six stages. Two of them run whether or not anyone remembers them, which makes this a guard rather than a checklist. Select a stage to see what it does.
Nothing is blocked or slowed down.
The difference comes earlier: the agent knows which parts of the change matter most before it starts, so you spend less time reviewing edits that were never going to work.
02 · The engine
You can call each one directly, and each writes to the same index the guard reads.
/epiclavis:indexAsk what a change will break, whether something already exists and what you noted last time.
/epiclavis:praevisioThe guard: foresight while you plan, and a check on every edit.
designKeeps a design system consistent by catching colours that drift from the palette, dark mode that falls behind and style rules that can never match.
dataKnows where personal data is stored, which routes are exposed and what a database migration would affect.
/epiclavis:writerTurns what the engine knows into documents people can read. The GDPR record of processing works today.
/epiclavis:initPoint it at a project and it takes stock of what is there. It takes one command to start.
An AI agent editing your codebase does not remember how the code fits together. A change can break something elsewhere, and whatever was learned about fragile modules is lost when the session ends. The next agent starts from scratch.
In regulated work the bigger problem is that the reasons behind decisions go unrecorded. Six months later, nobody can say why something was built the way it was. In pharma that decides whether you can defend a system to an inspector.
Praevisio is Latin for foresight. It checks twice: once while you plan, and again automatically on every edit. It doesn't guess, so the same question always gets the same answer.
It catches the knock-on effects of editing a shared component, unused code and passwords or keys left in source code. It also catches database calls that skip row-level security and colours that fail the WCAG accessibility contrast standard. Design tokens that have drifted from their definitions, and edits to files that should not change, are flagged too.
It also says when a file is missing from the index. An audit tool that can't see a file is silent, just like one that checked and found nothing. The first is far riskier. So the engine shows which checks are active for your project and which are not.
Acta replaces the to-do and issue files that tend to go stale in a repository. It tracks to-dos and issues. It also records what usually goes unrecorded: architectural decisions and the reasons for them, requirements through their lifecycle, conventions, and a log of what worked and what didn't.
Requirements move through draft, review, approval, implementation and verification. An approved record cannot be edited: a change creates a new version and retires the old one. Database triggers enforce this, which gives an auditor more assurance than a written procedure.
Notes are saved as the work happens, so nothing needs reconciling at the end of a session and you can close the terminal at any time.
Each feature is mapped to the frameworks it supports: ALCOA+, 21 CFR Part 11, EU Annex 11, GAMP 5 and GDPR. The decision log, for example, follows ALCOA's principles of attributable, contemporaneous, original and enduring records, and the audit-trail requirements of Part 11 and Annex 11.
Going past a governance warning is a deviation in GxP terms, and epiClavis records it as one. The audit trail then shows what was built, where governance was bypassed and who bypassed it.
Capture is always on and cannot be switched off, because an audit trail you can disable is not an audit trail. Each project chooses how much process to add on top, so the engine works inside and outside regulated settings.
Working today: code analysis for JavaScript, TypeScript and Python; Praevisio with all its checks; Acta with the full requirement lifecycle and approval lock; a health command that reports the engine's status as qualification evidence; and report generators, of which the GDPR Article 30 record of processing is the furthest along.
Designed but not yet built: a desktop workbench organised around the validation lifecycle, UAT capture that produces test evidence, a CAPA path for closing deviations, a traceability matrix and a change-control record generator. These are plans, not features.
The engine is the part you install. It runs as a plug-in and is meant to be free, so that any project can use it.
The Workbench is the part I plan to sell. It is the same engine with an interface that follows validation from planning to audit. It suits people who would rather see the traceability than search for it.
It isn't publicly available yet, so there is nothing to download. If you'd like to hear when it is, get in touch.