The serve runtime, all 14 wedges, every CLI command, every doc you’re reading — open source under BSL 1.1, free forever for personal, academic, and commercial use that doesn’t compete with FastCrest’s own hosted offering.
The Pro tier exists for one specific feature: continuous self-distillation. Reflex serves your model in production while quietly distilling a customer-specific 1-step student from your traffic, gating with a 9-check methodology, and hot-swapping when the new student passes. No other deployment tool ships this; we ship it because we own both the serve runtime and the distill pipeline.
Open source
$0 / forever
All 14 wedges. Full CLI. Local Pagefind search. ONNX export with cos = +1.0 parity. SnapFlow distillation. Discord support. Community Pro previews.
Includes continuous self-distillation OFFLINE — train a student manually, manually swap into production.
Pro
$99 / month per deployment
Everything in Open source, plus:
Includes 4 distill runs/month (Modal A100, ~$8-12 each) + 8 eval runs/month (Modal A10G, ~$1-4 each), with a $60/mo Modal spend cap.
Enterprise
Custom
Everything in Pro, plus:
Contact us — typically responds in 1-2 business days.
| Capability | Open source | Pro | Enterprise |
|---|---|---|---|
reflex go / reflex serve | yes | yes | yes |
| All 14 runtime wedges | yes | yes | yes |
reflex export with cos = +1.0 parity | yes | yes | yes |
| SnapFlow 1-step distillation | manual (via reflex train distill) | continuous loop | continuous loop |
| 9-gate eval methodology | — | yes | yes |
| Atomic warm-swap (policy versioning) | substrate only (single-policy circuit breaker) | full 2-slot rollback | full 2-slot rollback |
| 24h post-swap monitor + auto-rollback | — | yes | yes |
| Customer-data fine-tune | — | yes | yes (with custom training loops) |
| Weekly Pro report | — | CLI; email/Slack in P1.5 | CLI + email + Slack |
| HF Hub artifact storage | DIY | included | optional (or on-prem) |
| Reflex Compute Pack hardware bundle | — | — | Phase 2 deliverable; reach out for pre-orders |
| Reflex chat (GPT-5 Mini hosted proxy) | 100 calls/day free | unlimited | unlimited |
| Discord support | community | community + tagged @fastcrest | dedicated channel |
| Email response SLA | best-effort | 1 business day | negotiated |
| Uptime SLA | none | none | 99.5% / 99.9% / 99.99% (negotiated) |
| License | BSL 1.1 | BSL 1.1 | negotiated commercial license |
A typical Pro customer running one robot deployment:
| Line item | Monthly $ |
|---|---|
| Reflex Pro license | 99 |
| Modal: 4 distill runs × $10 | 40 |
| Modal: 8 eval runs × $2 | 16 |
| HF Hub: ~5 GB artifact storage | 0 (free tier) |
| Total run cost | ~$155 / mo |
You pay Modal directly via your own account; Reflex never proxies your compute. The Pro license unlocks the orchestration; the compute spend is at-cost.
Reflex’s source is licensed under BSL 1.1. The Pro license is a separate commercial agreement that grants access to the continuous-distill loop’s licensed binaries (Pro tier ships as a separate package extra). Cancelling Pro reverts your install to the open-source feature surface; the open-source code keeps working forever.
# Get a Pro license (Phase 1: dev license bound to current hardware;# Phase 1.5 moves to a real license server)python -c "from reflex.pro import issue_dev_license; issue_dev_license(customer_id='your-org')"
# Or contact us for the production license server# hello@fastcrest.comThe Pro feature gates --pro on the serve command. Without a valid license, the flag is rejected at startup with a clear error — no silent fallback.