INVESTORS

Private Apple-native intelligence
infrastructure on hardware you own.

EvoCortexAI builds private Apple-native intelligence infrastructure around user-owned Apple hardware — orchestrated into a coordinated private inference mesh. The fabric coordinates the work; the user owns the hardware, the data, and the intelligence layer. No centralized inference provider is required by default.

THE THESIS

Why this, why now.

◎
AI adoption is accelerating.
Organisations are deploying AI into sensitive workflows: health, legal, finance, personal communication. The default delivery model is a cloud API where data leaves the user's control.
⬡
The sovereignty segment is underserved.
A segment of users and organisations requires that inference and data remain on hardware they control. Apple Silicon has made this technically viable at a reasonable cost. No mature platform exists for it yet.
EvoIntelligenceFabric is the platform for that segment.
Saturn-Control (control plane), Saturn-Node (inference execution), and the Saturn One Apple app (client) form a coordinated system. The user provides the hardware; we provide the coordination layer. No cloud model API is required by default; deployed as documented, inference traffic stays within user-operated infrastructure.
EARLY PERFORMANCE INDICATORS

Measured on user-owned M4-class hardware (pre-seed review only).

All numbers are early controlled indicators. public_claim_allowed = false. Full repeatability protocol pending. See Benchmarks for charts, footnotes, and limitations.

Tier A (Sustained 30-min, complete evidence):
7B model — 54.2 output tokens/sec at 26.8 W wall power (≈2.02 tokens/sec/W)
Energy uses full run duration. Still early; N=1, 5-min wall power resolution. Repeat runs and complete artifacts required before public claims.
Tier B (Short/medium runs, partial evidence):
3B-class short interactive: 124.9 tok/s @ 19.4 W
7B-class short: 89.4 tok/s @ 23.1 W
7B-class medium reasoning: 61.0 tok/s @ 27.3 W
14B-class medium: 34.6 tok/s @ 30.1 W
All N=1, wall power 5-min resolution. See full limitations and source files on the Benchmarks page.
Baselines (30-min):
Saturn-Control (full stack): 3.5 W • Saturn-Node-01 idle: 3.0 W
CURRENT STATUS

What is built and operational today.

The following reflects the state as of mid-2026. No revenue, valuation, or projection figures are provided on this page.

Company EvoCortexAI S.L., incorporated in Spain
Saturn-Control Production — Swift/Vapor, PostgreSQL, native macOS LaunchDaemon
Saturn-Node Operational — MLX inference on Apple Silicon, connected to control plane
OpenAI-compatible gateway Operational — /v1/chat/completions proxied through Saturn-Control
Saturn One (Apple-native client) In development — not yet publicly available
Benchmarks Methodology prepared — first compliant measurements pending
GitLab (self-hosted) Operational — git.evocortex.ai
WHAT WE DO NOT CLAIM

Our commitment to accurate communication.

Investor materials that overstate traction or capability erode trust over time. We avoid that.

  • ✕  No revenue figures published here (not because there are none to share, but because this page is not a private placement document)
  • ✕  No user or customer count claims
  • ✕  No benchmark numbers until a methodology-compliant run is complete (see Benchmarks)
  • ✕  No efficiency comparisons to other hardware or cloud providers without identical-condition measurements
  • ✕  No valuation or projection claims
Serious investor inquiries welcome.

We respond to substantive inquiries with a brief written summary of the current state. No deck is available on request without a conversation first.

ceo@evocortex.ai
Dmitrii Sukhov · Founder · EvoCortexAI S.L.