How Virex improves over time — the running improvement story

Improvement in Virex is engineering work, not an abstract capability leap. Concretely, that means the classifier routes more prompts to the right template on the first try, builds waste fewer tokens, the Deep Engine picks the right lane per edit automatically, and the output lands closer to what the brief actually asked for.

Monthly releases are the visible milestones along that curve. Individual release notes live in the Updates category.

Live preview now works across demo, build, and edit / repair — every flow opens the generated repo in a real workspace instead of a static screenshot.

The Standard Engine handles cosmetic edits (colors, sizing, logo position, dark mode) unlimited on every plan. They do not touch the Deep Engine allowance.

Every edit request is classified as micro, small, medium, or large on the way in, and routed automatically to the cheapest lane that can do the job.

Template selection got tighter — the engine picks a better-fitting template on the first pass more often, so fewer builds need a retry.

Mobile continuation: Expo-aware routing, clearer foundation choices, better first-user mobile guidance in the docs.

Docs clarity: dedicated Build, Plans, Deploy, Scope, and Tools pages with cross-links instead of one long wall of guidance.

Deep Engine credit transparency: visible cap progress, soft-cap downgrades before the limit, and a clear fallback to the Standard Engine at 100%.

Stronger repo import for projects that did not start in Virex.

More deterministic template variants (so common product shapes — webshop, dashboard, marketplace — land cleaner each release).

Lightweight community and discovery layers once the core builder story is stable.

If you want the short version, the Updates category in Insights lists monthly release notes back to January 2026 — each a self-contained summary of what changed that month. If you want the rolling narrative, this page is the continuous version.