Open-core analyzer.
Pay for the learned passes.
The CLI, the web analyzer, the USDZ validator, and the public-repo GitHub Action are MIT-licensed and free forever. Pro adds the work the autonomous loop has been shipping: a learned similarity index, cohort-recipe auto-tune, LOD bundles, saliency-aware decimation, and per-asset visual-diff gating.
Looking for multi-tenant CI for a production agency? See agency plans →
The full standards-first analyzer + a baseline optimizer. MIT-licensed core.
- Unlimited analyze
- 50 optimize jobs / month
- web-mobile preset
- USDZ validator
- Public-repo GitHub Action
Everything in Free plus learned similarity, cohort auto-tune, LODs, and the optimization passes shipping from the autonomous loop.
- Unlimited optimize jobs
- All presets (web-mobile, web-desktop, quality-max)
- Similar-assets MLP (SPEC-0087)
- Cohort-recipe auto-tune (SPEC-0044)
- LOD bundles + saliency decimation
- Per-asset visual-diff gating
- Shopify admin texture dedup (SPEC-0091)
| Feature | Free | Pro |
|---|---|---|
| Analyze | ||
| Analyze (glTF / GLB / USDZ) | Unlimited | Unlimited |
| Optimize | ||
| Optimize jobs / month | 50 | Unlimited |
| Presets | web-mobile only | web-mobile · web-desktop · quality-max |
| Baseline passes (cohort-quantize + channel-pack + meshopt)SPEC-0079 | ✓ | ✓ |
| Compare | ||
| Compare assets side-by-side (orbit-8 SSIM + ΔE94 + verdict)SPEC-0093 | Sample demo only | ✓ |
| Batch compare N assets (up to 8) — N×N similarity matrixSPEC-0093 | 4-asset sample only | Up to 8 assets |
| USD / AR | ||
| USDZ parity validator (12+ Apple-AR checks)SPEC-0033 | ✓ | ✓ |
| USDC writer (binary USD instead of crate)SPEC-0076 | × | ✓ |
| Pro features | ||
| Similar assets (learned MLP, per-tenant embedding index)SPEC-0087 | native-v1 fallback | learned-v1 (mlp-v0) |
| Cohort-recipe auto-tune (-44% to -97% gz on the right asset class)SPEC-0044 | × | ✓ |
| Source-texture dedup batch (Shopify admin + CLI)SPEC-0091 | × | ✓ |
| LOD bundles (progressive high / medium / low patches)SPEC-0080 | × | ✓ |
| Saliency-aware decimation + saliency-driven texture budgetsSPEC-0048 | × | ✓ |
| Per-asset visual-diff gating (ΔE94 budget)SPEC-0006 | × | ✓ |
| Quick-View asset gallery (favorites + saved views)SPEC-0099 | Sample corpus only | Full history + favorites + saved views |
| Pro dashboard (recent jobs + favorites + activity heatmap)SPEC-0103 | Upgrade prompt only | ✓ |
| Distribution | ||
| GitHub Action — public reposSPEC-0012 | ✓ | ✓ |
| Private-repo GitHub Action + Slack delivery | × | Team add-on |
| Multi-tenant orgs | × | Studio add-on |
- Shopifynative app
- three.jsr170+
- <model-viewer>web
- Apple Vision ProUSDZ
- UnityglTF runtime
- UnrealglTF runtime
- GitHubAction + CI
- VS Codeextension
Everything we get asked twice
- No. Analyze is unlimited forever; optimize is capped at 50 jobs / month per tenant. Upload size limits come from the hosting layer (Vercel ~25 MB) not the tier.
- On Shopify-embedded installs we use Shopify's
appSubscriptionCreateflow (seeapps/shopify-app/lib/billing.ts). For direct API tenants Pro is enabled via theSPATIALPACK_PRO_TENANTSallowlist after a Stripe invoice — contact us to set one up. - Free queries use a hand-tuned 32-dim native-v1 embedding; Pro queries use mlp-v0, the trained MLP shipped with SPEC-0087 + SPEC-0089. Same neighbor API; materially better cosine separation on hard-shaded assets and on cohort-mixed catalogs.
- Anytime, from inside the Shopify app or by emailing us for direct deploys. Pro features stop at the end of the billing cycle; your existing job history and embedding indices stay readable on the Free tier.
- Per the 6-month-delayed-OSS convention, the optimization passes do. The learned-MLP weights, the cohort-recipe auto-tune index, and the multi-tenant infra stay paid.
- SSO, dedicated capacity, on-prem option, 99.95 % SLA — all live on the agency plans page.
Try Pro on your worst-offender asset.
The 14-day trial doesn't require payment up-front on Shopify. Drop in the GLB your designer keeps re-exporting and watch the cohort-recipe pass take it from 8 MB to the 800 KB target — without crushing the visual diff.