Investor Reference · Device Configuration Matrix

One architecture. Eight ways to live in it.

LeptonX is a zero-custody, on-device intelligence platform — and that principle holds across every device class from a $9,000 sovereign cluster down to a phone the user already owns. This matrix maps the full hardware spectrum, the services each tier supports, the subscription levels, and the customer profiles each tier serves best.

8
Tiers · $0 to $11,000
3×
Sovereign tiers (1–4)
2×
Sovereign Essentials (5a / 5b)
3×
Companion / Connect / Essential
The Matrix

Click any tier to expand its full configuration.

Highest-cost / highest-capability at the top, lightest-weight at the bottom. The Sovereign Essentials tiers (5a / 5b, in sage) are the consumer volume sweet spot — they deliver Maya voice, RAG, FHIR sync, and structured search standalone, with optional Spark pairing as an upgrade path.

Filter
Tier 01
Sovereign Plus
Pro Cluster
Dual NVIDIA DGX Spark GB10 ConnectX-linked; 256 GB unified $9,400 – $11,000
Family hub running Maya 35B + research-tier 235B, ComfyUI, LeptonX-Ray Vision, Z-Anatomy, Evo2-class genomics. Multi-patient (3–5).
$249–$399
/ month
Compute
GPU
2 × Grace Blackwell sm_121, CUDA 13
CPU
2 × 20-core ARM64 Grace
Memory
256 GB unified (128 GB × 2)
Storage
8–16 TB NVMe SSD
Link
ConnectX 200 Gb/s interconnect
Supported Services
  • Maya voice — Qwen3.5-35B-A3B + research-tier 235B (second-opinion mode)
  • ComfyUI / Flux / LTX-2 video & imagery
  • Z-Anatomy + Maya Anatomy Viewer
  • LeptonX-Ray Vision™ (DICOM → 3D segmentation)
  • Multi-patient registry (family of 3–5)
  • Genomic / DNA modeling (Evo2-7B class)
  • BonesRGood Verified tier (subset)
Key Use Cases
  • Caregiver of complex oncology / rare-disease patient
  • Family hub: parent + children on one device
  • Patient-led second-opinion synthesis from imaging + literature
  • DICOM segmentation + volumetric tracking over time
  • Voice + 3D anatomy navigation
Tradeoffs / Ideal Customer
  • Highest upfront cost in product line
  • Requires dedicated workspace; ~150–200 W under load
  • ARM64/sm_121 quirks (Blender CPU fallback, source builds)
  • Ideal: caregivers of medically complex loved ones, oncology / rare-disease households, self-directed health researchers
Tier 02
Sovereign
Reference Tier
NVIDIA DGX Spark GB10 The Maya patient unit; 128 GB unified $4,699
Full Maya pipeline. Qwen3.5-35B Q4 at ~45 tok/s. Multi-system FHIR sync, Z-Anatomy, LeptonX-Ray Vision, ComfyUI.
$129–$199
/ month
Compute
GPU
Grace Blackwell sm_121, CUDA 13
CPU
20-core ARM64 Grace
Memory
128 GB unified
Storage
4 TB NVMe SSD
Power
~170 W under load
Supported Services
  • Full Maya voice pipeline (Qwen3.5-35B MoE Q4, ~45 tok/s)
  • Multi-patient registry (1–3 active)
  • Particle Taxonomy® RAG (BGE-M3 + reranker + Qdrant)
  • FHIR multi-system sync (Epic, Scripps, MSK, UCSD)
  • Z-Anatomy 3D viewer + voice-to-region navigation
  • LeptonX-Ray Vision™ (single-organ segmentation)
  • BonesRGood Verified DEXA integration
Key Use Cases
  • Single patient with complex condition
  • Voice-first daily clinical Q&A on personal record
  • Imaging review with 3D body navigation
  • FHIR auto-sync from MyChart-class portal
  • On-device research literature retrieval
Tradeoffs / Ideal Customer
  • Single-GPU contention under combined workloads
  • Stationary device; not for mobile clinical encounters
  • Best with hardwired Ethernet or strong Wi-Fi 6
  • Ideal: sovereign-minded patients with one complex active condition, privacy-prioritized professionals, beta cohort
Tier 03
Sovereign-Lite
Apple Silicon
Mac Studio M4 Max 64 GB or M5 Max MacBook Pro 128 GB $2,200 – $4,500
Qwen3.5-30B-A3B Q4 via Metal. Single-patient FHIR. Browser-side anatomy viewer. Whisper-quiet, ~30–80 W.
$79–$129
/ month
Compute
GPU
Apple GPU 32–40 core, Metal/MPS
CPU
12–18 core Apple Silicon
Memory
64–128 GB unified, 614 GB/s on M5 Max
Storage
1–4 TB NVMe SSD
Power
~30–80 W
Supported Services
  • Maya voice (Qwen3.5-30B-A3B Q4 via llama.cpp + Metal)
  • Single-patient FHIR sync from MyChart
  • BGE-M3 embeddings on MPS
  • Web-based Maya Anatomy Viewer (browser WebGL)
  • ComfyUI image generation (Metal — slower than CUDA)
  • Document retrieval & voice Q&A
  • BonesRGood Enhanced + Verified
Key Use Cases
  • Privacy-prioritizing patient or caregiver, single condition
  • Voice + text Q&A on personal medical record
  • Family member managing aging parent's records
  • BonesRGood Verified with DEXA history
Tradeoffs / Ideal Customer
  • No CUDA — stack ported to Metal/MPS
  • Some CUDA-only research workloads unavailable (Evo2 class)
  • Limited multi-patient capacity (1–2 patients)
  • Ideal: Apple-loyal households, privacy-conscious knowledge workers (40–60), adult children managing parent's records
Tier 04
Sovereign-Core
Entry On-Device AI
Mac Mini M4 Pro 24–48 GB or compact Linux PC w/ 32 GB $1,099 – $1,799
Maya voice — Qwen3-14B Q4_K_M. Single-portal FHIR sync, RAG, voice Q&A, structured record search.
$39–$79
/ month
Compute
GPU
Apple M4 Pro GPU or integrated dGPU
CPU
10–14 core Apple Silicon / x86
Memory
24–48 GB unified
Storage
512 GB – 2 TB NVMe
Power
~20–35 W
Supported Services
  • Maya voice — Qwen3-14B Q4_K_M (or 30B at slower throughput)
  • Particle Taxonomy® RAG with bounded patient corpus
  • FHIR sync from one MyChart-class portal
  • Voice Q&A + structured record search
  • Web-based Maya Anatomy Viewer
  • BonesRGood Enhanced (full); Verified (DEXA)
  • No native ComfyUI / no DICOM segmentation
Key Use Cases
  • Beta tester wanting on-device AI under $2K
  • Single-patient FHIR sync + voice Q&A
  • Document retrieval, medication review, appointment summaries
  • Family caregiver with one or two patients
Tradeoffs / Ideal Customer
  • Cannot run Maya + ComfyUI + DICOM segmentation simultaneously
  • Smaller LLM may need stronger RAG support to match 35B quality
  • Voice latency 2–4 s vs. ~1–1.5 s on Tier 2
  • Ideal: cost-sensitive sovereign patient (~$1K–$2K), adults 50–70 with one chronic condition, BonesRGood Verified power-user
Tier 5a
Sovereign Essentials
Static · Mini PC
Intel N305 / Ryzen AI 5 16 GB DDR5 mini PC + 512 GB NVMe $350 – $500
Standalone-capable. Summary-grade Maya (Qwen3-3B / Phi-3.5), full RAG, FHIR sync, structured search. Optional Spark pairing.
$29–$49
/ month
Compute
GPU
Integrated iGPU + 11–50 TOPS NPU
CPU
8-core x86 (N305) or 10-core ARM (M2)
Memory
16 GB DDR5 (or 8 GB unified)
Storage
512 GB NVMe SSD
Power
~10–25 W
Supported Services
  • Maya voice — summarization-grade (Qwen3-3B / Phi-3.5-mini Q4)
  • Particle Taxonomy® RAG over bounded patient corpus
  • FHIR sync from one MyChart-class portal (full)
  • Structured record search (SQLite)
  • Lifelong access to records previously synced
  • BonesRGood Enhanced + Verified
  • Optional Spark pairing for heavy queries
Key Use Cases
  • Cost-conscious sovereign patient with bounded condition
  • Daily voice Q&A (summary-grade) on personal record
  • Lab review, medication check, appointment prep
  • Standalone household — no Spark required
  • Optional companion to a household Spark (Mode 2)
Tradeoffs / Ideal Customer
  • Voice Q&A is summary-grade — not Maya-35B depth
  • Voice latency 3–5 s on first token
  • Complex synthesis queries gracefully defer
  • 16 GB ceiling caps simultaneous workloads
  • Ideal: retiree managing one MyChart, BonesRGood Verified user, $400 hardware budget
Tier 5b
Sovereign Essentials
Companion · Mobile
User's existing phone or tablet iPhone 15+ / iPad / Pixel 9+ / Galaxy S24+ $0 (already owned)
The consumer sweet spot. Apple Foundation / Gemini Nano local LLM. Standalone or paired. The largest segment by volume.
$19.99–$29.99
/ month
Compute
NPU
A18 Pro / Snapdragon 8 Elite
CPU
6–8 core mobile
Memory
8–16 GB
Storage
256 GB – 1 TB
LLM
Apple Foundation / Gemini Nano + 3B fallback
Supported Services
  • Maya voice via Apple Foundation Models / Gemini Nano
  • Particle Taxonomy® RAG over phone-resident corpus
  • FHIR sync from one MyChart-class portal
  • Structured search + voice Q&A
  • Wearable integrations (Apple Health, Fitbit, Whoop, Oura, Garmin)
  • BonesRGood Verified + Companion features
  • Optional pairing with household Spark over Tailscale
Key Use Cases
  • BonesRGood Verified subscriber wanting Maya basics
  • Mobile-first sovereign experience — no second device needed
  • Family member at work pairing with home Spark for hard queries
  • Daily voice check-ins, medication confirmations
  • Travel-friendly access to lifelong records
Tradeoffs / Ideal Customer
  • Standalone capable; voice is summary-grade (3B-class)
  • Heavy queries route to paired Spark or queue
  • Battery use elevated when running local model
  • Ideal: BonesRGood demographic 25–45, mobile-first user, spouse of Tier 2 owner. Largest expected segment by volume.
Tier 06
Companion
Mobile Premium
iPhone 16/17 Pro or premium Android Tethered to home Spark over Tailscale $999 – $1,599
BonesRGood Verified plus Maya tethered to a household Spark. Wearable + clinical data fusion. Mobile companion to Tier 1–4.
$19.99–$39.99
/ month
Compute
NPU
Apple A18 Pro / Snapdragon 8 Elite
CPU
6–8 core mobile
Memory
8–16 GB
Storage
256 GB – 1 TB
LLM
3–7B local, full Maya via paired Spark
Supported Services
  • BonesRGood Verified (full)
  • Maya Companion App tethered to home Spark
  • Local FHIR record viewer (encrypted at rest)
  • MyChart sync via OAuth2 PKCE wizard
  • Wearable integrations (Apple Health, Fitbit, Whoop, Oura, Garmin)
  • Notifications, re-auth prompts, structural reports
Key Use Cases
  • Daily mobile access to your medical record
  • Wearable + clinical data fusion (Structural Score)
  • Maya voice-on-the-go (talks to home device)
  • DEXA scheduling, lab review, appointment prep
Tradeoffs / Ideal Customer
  • Best with paired home Spark for full Maya experience
  • Imaging segmentation, Z-Anatomy, ComfyUI unavailable locally
  • Subscription gates ongoing FHIR sync
  • Ideal: existing LeptonX household member with mobile access need; BonesRGood demographic 25–45
Tier 07
Connect
Mainstream Mobile
iPhone 14/15 or mid-range Android Structured viewer + BonesRGood Enhanced $400 – $800
Portal-connected BonesRGood Enhanced. MyChart sync. Structured record viewer. No on-device LLM.
$9.99–$19.99
/ month
Compute
GPU
Mid-tier mobile (no on-device LLM)
CPU
4–8 core mobile
Memory
4–8 GB
Storage
128–256 GB
Network
LTE/5G + Wi-Fi
Supported Services
  • BonesRGood Enhanced (portal-connected)
  • MyChart sync via FHIR OAuth2 PKCE
  • Structured medical-record viewer
  • Wearable integration (Apple Health, Google Fit, Fitbit)
  • Structural Score (Enhanced — labs + activity)
  • Read-only access to Spark-synced records
Key Use Cases
  • BonesRGood subscriber tracking Structural Score
  • User who wants MyChart organized in one place
  • Patient who wants secure mobile viewing without a Spark
  • Onboarding step before purchasing a sovereign device
Tradeoffs / Ideal Customer
  • No on-device LLM — Maya queries unavailable unless paired
  • No 3D anatomy, no LeptonX-Ray Vision, no DICOM
  • Imaging files exceed practical mobile storage
  • Ideal: mainstream consumer 25–45, athlete tracking longitudinal metrics, future graduation candidate
Tier 08
Essential
Free / Web Access
Any modern smartphone or browser BonesRGood Estimated tier — entry funnel $0 – $400
Free Structural Score. Wearable integration. Educational content. Top-of-funnel for upgrade to higher tiers.
Free
or $4.99 ad-free
Compute
GPU
Whatever the device has
CPU
Any modern mobile/desktop SoC
Memory
≥ 3 GB
Storage
≥ 32 GB
Network
Wi-Fi or cellular
Supported Services
  • BonesRGood Estimated tier — free Structural Score
  • Wearable integration
  • Bone-loading minutes tracker
  • Educational content
  • Account creation and onboarding for higher tiers
  • MyChart export upload (read-only summary)
Key Use Cases
  • First-time visitor exploring BonesRGood
  • Trial of basic skeletal health tracking
  • Onboarding funnel for upgrade to higher tiers
  • Educational engagement around bone/muscle health
Tradeoffs / Ideal Customer
  • No clinical record integration
  • No Maya AI features
  • Structural Score is regression-estimated, not clinically verified
  • Ideal: top-of-funnel — anyone 18+ with a phone, free-tier holdouts, international users where higher-tier hardware unavailable
Sovereign Essentials · 5a / 5b

Three operating modes. No Spark required.

Tiers 5a and 5b are explicitly architected to work standalone. They become more powerful if a household adds a Spark, but they don't require it. This is the consumer sweet spot — the volume tier that drives adoption.

Mode 1

Standalone — no Spark anywhere

The device runs entirely on its own. FHIR sync, RAG over a bounded patient corpus, structured record search via SQLite, and summarization-grade voice Q&A using a 3B-class model. Bounded queries — "what was my last A1C," "summarize my last three visits," "when's my next appointment" — are well-served. Complex synthesis is gracefully deferred.

≈ 80% of Customer Base
Mode 2

Paired with a household Spark

If a household already owns a Tier 1–4 device, the small device pairs with it over Tailscale. Heavy queries it can't handle stream to the Spark and back. The pairing is asymmetric (the small device works fine alone) and supports real family scenarios — spouse with the Spark at home, partner carrying 5b at work; or grandkid managing grandparent's Spark from their own mini PC.

Optional Upgrade Path
Mode 3

Second-opinion cloud lane

For users without a Spark who occasionally want deep analysis, a future opt-in mode would send a structured query plus minimum-necessary, de-identified context to a LeptonX-hosted endpoint. Per-query consent, on-device PHI minimization, rate limits. Gated on HIPAA counsel review (same gate as FPIN telemetry). Not required to ship.

Opt-in · Post-Counsel
Honest Notes

Assumptions, caveats, and where the matrix bends.

Investor-grade transparency. Every tier in the matrix carries assumptions about hardware availability, software compatibility, and pricing windows. Here is the honest accounting.

Pricing

Hardware costs reflect early-2026 retail.

The DGX Spark moved from $3,999 to $4,699 due to memory supply pressure; we hold $4,699 as the planning figure. Apple Silicon prices are taken from current Apple US store; M5 Max upper-tier configurations are estimates based on prior product cycles. Tier 5a hardware (mini PCs) is at retail price as of May 2026.

Tier 5a / 5b

"Summarization-grade" Maya is a real category.

On 16 GB DDR5 hardware, Maya runs as a 3B-class quantized model at 4–7 tok/s with 3–5 s first-token latency. Sufficient for: "what was my last A1C," "summarize my last visits," "explain this lab result." Not sufficient for: cross-document synthesis, multi-modal imaging interpretation, longitudinal trend analysis. Those defer to a paired Spark.

Tier 4 future

Sovereign-Core is a moving target.

The current entry point is ~$1,099–$1,799 (Mac Mini M4 Pro). Plan-of-record target is $700–$900 within 18–24 months as Apple M5 lands and M4 prices compress on the secondary market. Model efficiency (Qwen-3-class MoE at Q4) is moving in our favor on the same curve.

Software constraints

The DGX Spark sm_121 reality.

The Spark's GB10 (sm_121, CUDA 13, ARM64) is bleeding-edge silicon. PyTorch, vLLM, llama.cpp, ComfyUI, BGE-M3 all run on GPU. Blender Cycles, Flash Attention 2, and PyTorch3D do not — these fall back to CPU or require source builds. Any new install must be GPU-compatibility-verified before recommendation.

Subscription

Pricing is illustrative, not committed.

Pricing in the matrix is for strategic positioning. Final pricing will be calibrated against beta feedback, manufacturing costs, support overhead, and the architectural principle that subscription gates sync, not access — records previously synced remain accessible offline forever, even with subscription paused.

Regulatory

Counsel-gated items are clearly marked.

BonesRGood's HIPAA BAA-exemption position has been informally validated by Manatt Health (Randi Seigel); state-privacy review with Stacey Gulick (Garfunkel Wild) is queued. The FPIN telemetry pipeline operates in local-only mode pending formal HIPAA counsel opinion. Tier 5a/5b Mode 3 is similarly counsel-gated.

One architecture. Eight ways to live in it. The Sovereign tiers prove the methodology; the Essentials tiers deliver the volume; the Companion tiers extend the surface area. The principle — zero data custody, lifelong access — is the same at every price point.
— LeptonX Strategic Reference, May 2026
LeptonX is a privacy technology company, not a healthcare provider. Maya and the LeptonX product suite are personal data navigators — voice-enabled companions that help you make sense of your own information and prepare for conversations with your physician. They are not medical devices. They do not diagnose, treat, or recommend.

The most important conversations about your health happen between you and your care team. Always consult your physician or qualified healthcare provider before making any health-related decision.

Want to see the matrix in motion?

The investor deck walks through hardware economics, subscription unit economics, and the "subscription gates sync, not access" architectural principle in detail.

Strategic Pricing → Request Investor Deck → ← Back to Investors