<\!DOCTYPE html> Silicate — The AI Kernel Operating System
Ring-0 AI Kernel Operating System

The Windows
of AI.

Silicate Zero is the first operating system where AI is the kernel — not an application running on top of Linux. Ring-0 execution. Bare metal. No syscall overhead. 149.3 tok/s on AMD EPYC.

Get Silicate Zero — Free How it works →
149.3
tok / sec
CPU-only on AMD EPYC 9354P
10×
faster
than typical llama.cpp desktop
0
Linux
bare metal Ring-0 execution
Silicate Zero — Ring-0 AI Kernel Architecture
Ring-0 execution
The Problem

For 50 years, operating systems were built for humans typing at terminals.

The architecture we inherited — kernel/user space splits, syscall overhead, process isolation — was designed for a different era. Running AI on Linux is like racing a Formula 1 car on a dirt road. The car is fast. The road is the bottleneck. Silicate Zero is the road built for the car.

Legacy Stack
Application Layer
Host OS (Linux / Windows)
Virtualization / Hypervisor
Hardware
4 context switches per inference call
Silicate Zero
AI Kernel (Ring-0)
Hardware Abstraction Layer
Bare Metal
Zero syscall overhead
Architecture

No user-space. No kernel/user split.
No process boundaries. No syscalls.

<\!-- Light-theme architecture SVG --> <\!-- Background --> <\!-- Ring-3 box --> RING-3 User Space Sandboxed Agents AgentLang Runtime App Binaries (AOT) Type-safe via Validator + Capability system <\!-- Arrow Ring3 → Ring0 --> <\!-- Ring-0 box (main kernel) --> RING-0 AI Kernel — The OS Core <\!-- Boot-LLM sub-box --> Boot-LLM Qwen3-1.7B Q4_K_M 149.3 tok/s (CPU-only) x86_64 + ARM64 deterministic <\!-- Validator sub-box --> Validator Type-safe code execution 270+ test cases passing No runtime JIT overhead <\!-- Bottom row kernel sub-components --> Arena Allocators O(1) bump-pointer Zero-Copy IPC Capability-typed handles AOT Codegen Native x86_64 <\!-- HAL strip --> Hardware Abstraction Layer — x86_64 · ARM64 · HVF <\!-- HAL down-arrow -->
Safety enforced by the Validator and type system — not page tables or hardware rings
Single Ring-0 address space — no kernel/user split, no context switch cost
Unikernel architecture — one address space, one purpose: AI inference at hardware speed
Product Family

The AI kernel OS stack.
One platform. Every layer covered.

Silicate Zero is the Ring-0 foundation — 100% free, AGPLv3, forever. Compute is the LLM engine built in. Add Nefesh, Cloud, or Benchmark as your needs grow.

Core Platform
Free · AGPLv3
Silicate Zero
OS Layer — Ring-0 AI kernel

The unikernel foundation everything else runs on. Zero syscall overhead. No host OS between AI and hardware. Validated at 149.3 tok/s on AMD EPYC 9354P CPU-only.

Zero Basic Free Everyday local AI OS. Consumer PCs, laptops, EPYC hardware.
Zero Bare Free Naked Ring-0 kernel. No POSIX layer. For OS researchers.
Zero Server Free → Series A EPYC/Xeon-optimised. SMP, native NICs, SLA, priority support.
Included with Zero
Silicate Compute
AI Layer — LLM inference engine

The LLM engine built into the kernel. No setup. No host OS layer. Routes Qwen3, Llama, Mistral and frontier models directly through Zero. This is the main event.

Free
Compute Standard
Always included
Built-in inference routing for every Silicate Zero installation. Qwen3 Q4_K_M at 149.3 tok/s on EPYC. Zero setup — it's the kernel.
$/month
Compute Premium
Dedicated inference
Dedicated high-end inference on your hardware. Anthropic, OpenAI, and Gemini models routed through Zero without a host OS layer.
Extensions
Free + Paid
Silicate Nefesh
Biometric Layer — physiological intelligence from sensors. Real-time body state (rPPG video or wearable) fed into kernel context.
BYOS API AGPLv3 · Free
Bring your own sensor. Plug in any rPPG stream or wearable. Silicate handles signal processing and biometric inference.
Agent M2M $/month per agent
Autonomous agents connect via MCP or llms.txt. Real-time biometric context fed directly into agent reasoning.
Paid
Silicate Cloud
Server rental. Zero + Compute + Nefesh bootable directly from the website. Managed infra, zero config, BYO model.
$/month hosting
Free + Paid
Silicate Benchmark
Performance Suite. Self-host CLI or rent dedicated test hardware for a verified results report.
Free CLI · Hardware rental paid
Enterprise
Silicate Vault
Certified Editions for regulated industries. Government / Defense (air-gapped, STIG), Finance / HFT (nanosecond timing), Healthcare (HIPAA + BAA).
Custom contract
Zero Basic and Zero Bare are 100% free, forever. The free tier ships complete — it is not a trial. Silicate's startup program offers free Zero Server until Series A for qualified startups.
Benchmark

Real hardware.
Real numbers. Not a slide deck.

Silicate Zero Server — AMD EPYC 9354P
64 logical CPUs · Native NIC · SMP enabled · No GPU
149.3tok/s
Qwen3-1.7B Q4_K_M — 32-token run, CPU-only, Ring-0
System / Runtime
Model
Mode
Result
Silicate Zero Server — AMD EPYC 9354P
Qwen3-1.7B Q4_K_M
CPU-only, Ring-0
149.3 tok/s
Fast desktop CPU, llama.cpp / Ollama class
Qwen3-1.7B Q4_K_M
CPU-only
10–15 tok/s
Intel i7-13700K, llama.cpp Q4
Qwen3-1.7B
CPU-only
35 tok/s
Consumer GPU, 6 GB VRAM
Qwen3-1.7B
GPU
~126 tok/s
Qwen official SGLang benchmark, 1 GPU
Qwen3-1.7B BF16
GPU serving
227.80 tok/s

GPU acceleration is the next target. The architecture doesn't exclude it — it delays it until the Ring-0 foundation is solid. Stage 16+.

We don't sandbox the AI inside a guest.
The AI is the kernel.
Roadmap
Stage 11 — Now
Boot-LLM Operational
Cross-platform deterministic. Validated on AMD EPYC 9354P. CPU-only inference at 149.3 tok/s. AGPLv3 open source.
Stage 12
Sandboxing + AgentLang
Full sandbox isolation between agents. AgentLang migration from experimental to production runtime. Type-safe agent execution.
Stage 14
Dual Desktop — Human Oversight
Classical user desktop alongside Agent Desktop: live visualization of agent reasoning, state, and intent. Pillar 2: Human Oversight at OS level.
Stage 24+
Pure-AgentLang Kernel
The AI kernel written in AgentLang itself. Self-hosting. The OS that runs itself. The logical endpoint of what we started.
Open Source

The kernel is free.
It stays free.

Silicate Zero Basic and Zero Bare are AGPLv3. The full kernel source is published openly. The stack — Zero, Compute, Nefesh — is designed to stay open. Silicate's startup program gives free Zero Server access until Series A — because the ecosystem matters more than early revenue.

Star us on GitHub GitHub stars
The Full Stack
Zero Basic
Local AI OS. Individuals.
Free · AGPLv3
Zero Bare
Pure kernel primitives. OS researchers.
Free · AGPLv3
Zero Server
EPYC/Xeon. Free until Series A.
Free → Series A
Compute Standard
Built-in inference. Every Zero user.
Free
Compute Premium
Frontier models on your hardware.
$/month
Nefesh BYOS API
Biometrics from any sensor.
Free · AGPLv3
Nefesh Agent M2M
MCP / llms.txt for autonomous agents.
$/month per agent
Silicate Cloud
Pre-installed stack. Managed infra.
Paid

The next operating
system has already started.

Silicate Zero is not a roadmap. It is shipped code running on real server hardware. The kernel is open. The vision is clear. The AI era needs a new OS — and it's here.

Get Silicate Zero on GitHub
Stage 11 of 24+ — and accelerating