AI review
Cadre’s AI reviewer reads your actual model — the geometry the kernel holds, not a screenshot — and flags the things that ruin a print, grounding the measurable ones in a checkable number and proposing a concrete fix.
Manufacturability flags
The reviewer surfaces issues as clickable markers that frame the problem in the viewport:
- Thin walls — a wall too thin to survive the process.
- Steep overhangs — surfaces that will sag or need support in FDM.
- Sharp internal corners — stress risers and hard-to-machine features.
- Small features — details below what the process can hold.
- Enclosed voids — sealed internal cavities that would trap uncured resin or unfused powder and cannot drain.
- Non-watertight solids — non-manifold geometry that will not print cleanly.
Grounded in measured geometry
This is the property that separates Cadre from a screenshot reviewer. A flag of a measured kind — thin wall, overhang, enclosed void, non-watertight solid — must cite the kernel’s own measured value. If the measurement does not support the flag, the flag is dropped before it reaches you.
So a thin-wall flag reads “wall 0.8 mm, threshold 2.0 mm” — a number you can check against a caliper — not a vague “looks thin.” See Methodology for how to verify this yourself.
Parameter fixes
For a part you modeled in Cadre, the reviewer proposes the fix as a typed
parameter change shown as old → new (for example, “wall 0.8 → 2.1 mm”). Accept
it and it applies as a single reversible edit through the
review inbox. For an imported mesh with no feature tree,
you still get the measured flag and the number to fix by hand.
Process presets
Choose a material and process — FDM, SLA, machined, or sheet metal — and the relevant checks and thresholds apply. A thin wall for one process is fine for another; the presets keep the flags honest to how you will make the part.
Honest scope
The reviewer explains, prioritizes, and suggests. Detection and grounding are done by the kernel; the AI turns measured numbers into plain-language flags and parameter suggestions. It is not a generate-a-part-from-text tool, and it never edits your model on its own — it proposes, and you accept.