The Flagship PlatformIn active development

ZylForge: The Engineering Operating System

ZylForge is not a CAD program. It is an Engineering Operating System built around a native parametric B-Rep kernel — modelling, validation and learning guided by deterministic checks instead of unconstrained AI output. It is in active development: no released build is downloadable from this site, and what follows separates what is implemented and tested from what remains architecture.

FEATURED DEMONSTRATION — ZylForge Ecosystem Knowledge Flow
JULY 23, 2026

Recorded demonstration — not a live production run.

The Problem

Why Traditional CAD Toolchains Are Fragmented

Engineering teams traditionally struggle with fragmented workflows: parametric modeling in Fusion, manual 2D drawing sheet creation, disconnected G-code preparation, and manual rendering in external software.

Generic AI helpers make matters worse by outputting unconstrained text or synthetic meshes that lack B-Rep geometry and fail in production.

The ZylForge Architecture Solution

Deterministic Physics over Unconstrained AI
Deterministic Learning Studio — Lessons Validated Against Real Application State
Deep B-Rep Geometry & Closed DXF Contour Verifiers
Cryptographic Merkle Root Release Certificates

Platform Subsystems

The Core Services of ZylForge v2

The subsystem architecture the v2 platform is being built against — a plan, not a list of finished features. The sections below separate what is implemented and tested from what remains design.

1. Workflow Engine

Sequences engineering jobs into resumable task queues with persistent state tracking across execution steps.

2. Capability Registry

Designed to track engine limits declaratively and reroute unsupported steps to a capable engine instead of failing the run.

3. Artifact Manager

Computes SHA-256 hashes, tracks file byte sizes, and guarantees zero 0-byte orphan files.

4. Verification Engine

Parses STEP B-Rep volumes, DXF closed loops, and CSV BOM items against the modelled intent.

5. Release Manager

Signs cryptographic Merkle root certificates (`RELEASE_CERTIFICATE.json`) and locks production manufacturing packages.

6. SQLite Engineering DB

Stores relational records for Projects, Assemblies, Materials, Parts, and Evidence Logs.

New Milestone

Learning Studio

The Learning Studio is a newly implemented milestone inside ZylForge's Education area: a deterministic lesson engine that guides a learner through an engineering task and validates completion against the learner's own sketch document — the real application state, not a video they watched or a button they clicked. The first studio lesson, Your First Sketch: Two Concentric Circles, is in place.

Lessons are data, driven by a state machine

Each lesson declares ordered steps — explanation, demonstration, action and validation — run by an explicit engine. Validation receives an injected snapshot of the modelling document, so lesson logic stays decoupled from any particular document, and the engine never mutates geometry itself.

Pause, resume, replay, exit

Pause keeps the session; resume continues from the same step. Individual steps and demonstrations can be replayed, and a lesson can be exited cleanly at any point.

Hints after failed attempts

When an attempt fails, the step surfaces hints after a set number of tries instead of blocking the learner — incorrect work opens a path, not a dead end.

Guided SVG demonstrations

Procedural SVG animations demonstrate each step beside the canvas. They are guided demonstrations, not live CAD operations — the real work happens in the sketch canvas. With reduced-motion enabled, the final frame is shown immediately.

Narration with visible text

Speech synthesis reads instructions aloud where the browser supports it, with pause and resume — and the instruction text is always visible on screen, so speech is never the only channel.

Accessible by construction

Keyboard-operable controls throughout, a labelled progress bar, and live-region announcements for phase changes, feedback and hints.

Education and assignments

How the studio fits the workflow

The Learning Studio sits inside ZylForge's existing Education area. A lesson catalogue with stable identifiers is what lecturer-side assignment references; the student assignment panel reads the same catalogue, so an assignment points at the real lesson rather than a duplicate list.

Lesson progress is recorded as the learner moves — opened, steps reached, validation passed, lesson complete — each state written when it happens.

Stated limits

Progress is device-local

  • Stored on the learner's own machine — not synchronised to a central server.
  • No cohort or institution-wide progress reporting exists in this build.
  • No institutional deployment or authentication has been independently verified.
  • Those capabilities belong to an institution layer that does not exist here yet.

Recorded honestly rather than implied: the Learning Studio is a milestone inside ZylForge — not a released Education product.

Development and validation evidence

What was checked on 19 October 2026

Each card below says what ran, where it ran, and what it covers. Commit under test: aec2b17 (main).

Vitest: 1,099 / 1,099

1,099 tests across 124 files, all passing. Desktop application TypeScript test suite — re-run independently for this page.

TypeScript: 0 errors

Type-check completed with a clean exit (tsc --noEmit) — re-run independently for this page.

Hosted CI: green

GitHub Actions CI passed on main at aec2b17 in 3 min 15 s on 19 October 2026.

Windows Production Build: green

Same commit, 1 h 2 min 43 s: workspace-wide Rust checks and tests, lint, application tests, production build, installer and MSI artifacts uploaded.

What this does and does not show

The Vitest figure covers the desktop application's TypeScript tests; the Windows production build additionally runs the Rust workspace checks and tests and produces the installer artifacts. Kernel, Rust, desktop and installer validation are kept distinguished here rather than merged into a single headline number. Passing build and CI signals shows the code compiles and its tests pass — it is not runtime proof that every feature has been exercised in the real world. Results outside these two verifications (for example earlier kernel suite counts) are owner-reported and stay in internal reports; they are not restated here as independent verification.

Current roadmap

Wave D2 — advanced part modelling

Everything in this section is ongoing development, not a delivered capability — appearing in a roadmap does not make something an available feature.

Shipped at UI_WIRED

Four verticals delivered between 12 and 19 October: slots and polygons; wedge with sweep cut and loft cut; rib and web; helical thread ridge — wired into the user interface, with runtime observation still pending.

Remaining — Vertical E: split

Split functionality, with a multi-body output architecture gap still to be solved before it can land.

Remaining — Vertical F: path pattern and reordering

Instances along a path, plus dependency-valid reordering of features — both still in progress.

Remaining — Vertical G: direct modelling

Face-move and offset operations — the last and hardest step of the wave, not yet delivered.

Open acceptance gate — D1.5 persistence

Awaits a manual persistence retest by the owner. The gate is not complete, and this page does not present it as closed.

Learning across Zylvex

Where the Learning Studio fits

The studio is learning-by-doing inside the engineering workbench: the learner works on real sketch documents and completion is validated by the real engine. The pages below are related learning surfaces — related in purpose, not the same system. ZylForge remains an engineering product; the studio is a milestone within it, not a standalone or released Education product.

Local ZylForge Technical Assistant

Explainable AI Assistant — Antigravity Philosophy for Physical Engineering