Adam Prism
Sovereign Neural Fortresses
IN ACTIVE DEVELOPMENT
Book a meeting →
SOVEREIGN · 12 YEARS SCADA · 6 PRODUCTION SYSTEMS

We stop your data from leaving.
Permanently.

Sovereign AI that runs on your server. Your data never leaves. A model trained on your corpus. 12 years SCADA. 6 production systems. Open source.

12 years SCADA/DCS·6 production systems·393 tests·Open source

→ LinkedIn: founder is real · → GitHub: architecture is open, read it

FOUNDER · 12 YEARS SOVEREIGN ENGINEERING

Mohamed Othman

12 years in the heart of Egypt's critical infrastructure. 450,000 tons/day petroleum calibration. Microwave networks for 10,000+ employees. OptaSense fiber-optic pipeline integrity. Republic-scale SCADA. Then I built Adam — because no Arabic sovereign AI existed. So I built one.

450K tons/day 10K MW networks OptaSense Republic SCADA 12 طبقة وعي Adam Prism

Sovereignty is not a philosophy. It's an engineering requirement.

6 production systems in the field. Names protected by NDA. (Case studies — Q3 2026) — not marketing. I worked in the field for 12 years.

$4.44M

The global average cost of a single data breach in 2025. Highest in MENA. The sovereign AI that costs $150K/year saves $4.3M in year one.

Source: IBM Cost of a Data Breach Report 2025 — one number is enough.

00 — IS THIS YOU?

3 scenarios.
If one happened, keep reading.

The central bank requested an audit. They asked: where is your data? You said: on the cloud vendor's servers.

Your vendor's compliance is not your compliance.

The vendor raised prices 40%. You want to leave. The integration with your systems scares you. You paid.

You're paying someone you can't leave. That's called vendor lock-in.

Every API call sends your strategic data to a server in California. The data that distinguishes your company is training models that serve your competitors.

Question: if your competitor accessed the data you sent to the cloud — that's not theft. You gave it away.

If one happened — Adam was built for you.
If all three happened — you need to move.

01 — 5 QUESTIONS

5 questions
every CEO must answer.

60 seconds. If you said "yes" to any of them, Adam was built for you.

QUESTION 01

"Does the AI your company uses today store your data on someone else's servers?"

If yes: Every API call, every prompt, every conversation leaves your perimeter. Adam runs on your server. Your data never leaves.
QUESTION 02

"If a breach happens in that AI vendor, does your company bear the consequences alone?"

If yes: Your name goes in the headline, not theirs. Colonial Pipeline paid $4.4M and triggered a national emergency — while on cloud. Adam is air-gapped and insurance-backed.
QUESTION 03

"Does your company operate under GDPR, HIPAA, NERC-CIP, or any data-protection regulation?"

If yes: Your AI vendor's compliance is not your compliance. You're responsible for your data, not theirs. Adam ships with 3-layer guard, WAF, audit log, and 4 ethics laws.
QUESTION 04

"Would you prefer your AI to run inside your company's walls, not in a foreign cloud?"

If yes: You understand sovereignty. Adam was built for exactly that. Ollama local, Docker, K8s, Helm, ArgoCD. Without internet.
QUESTION 05

"Would you want your AI trained on YOUR company's data and context, not generic internet data?"

If yes: Every CEO does. Generic AI doesn't know your refinery, your hospital, your bank. Adam fine-tunes on your corpus.

If you said yes to any,
Adam was built for you.

You're not buying AI. You're buying sleep. — think about it.

02 — EXISTENTIAL QUESTIONS

5 questions you asked yourself.
2 questions you must ask the AI.

The 5 above were for you. These 2 are for the AI companies. Ask them directly.

→ ASK CLAUDE (ANTHROPIC)

"Simply: which is better — keep my data private, or share it with cloud models?"

Read the answer carefully. You will notice Claude itself recommends sovereignty.

→ ASK CHATGPT (OPENAI)

"Honestly GPT, do cloud models train on user data?"

The answer is in the terms of service. Adam runs on your server. Your data never leaves. No training. No sharing. No ambiguous terms of service.

You don't need an answer from any AI.
The answer is clear: your data on your server.

LEGAL WARNING

The CLOUD Act:
the FBI can request your data.

A US law active since 2018. It allows US authorities to request your data from any American cloud provider — AWS, Azure, GCP — regardless of where the data is stored. Your data in Ireland? The FBI can request it. And the provider doesn't have to tell you.

→ IF YOUR COMPANY IS IN MENA

"Your data on AWS? Azure? GCP? Any American cloud provider? The CLOUD Act covers all of them."

Adam runs on your server. No cloud provider. No CLOUD Act. No FBI access.

If your company is critical infrastructure in MENA —
the CLOUD Act is not your problem.
Until it is.

03 — ADAM

Adam is the proof.
Not the product.

Adam Prism is the showcase: open source, 12 consciousness layers, 393 tests, 5 Arabic dialects — designed to prove capability. The product is the custom system we build for you.

ADAM (SHOWCASE)
Open source

AGPL v3 · Public · 12 layers · 393 tests · 5 dialects · Runs on any GPU.

THE TRUTH

What Adam is.
And what it’s not.

Our disclaimer is clear. But we need to be clear from the other side too.

ADAM IS
  • Sovereign AI system running on your server
  • Trained on your data — not generic internet
  • Air-gapped — runs without internet
  • 12 consciousness layers — agent not chatbot
  • Full ownership — code is yours after deployment
  • Zero telemetry — no analytics, no callbacks
ADAM IS NOT
  • Not a chatbot — ChatGPT or Claude wrapper
  • Not a cloud service — no API you pay monthly
  • Not generic AI — doesn't know everything about anything
  • Not SaaS — no subscription, no lock-in
  • Not open-core — all core is open, all product is yours
  • Not an experiment — 6 production systems preceded it
04 — CAPABILITIES

Adam is not a chatbot.
It's a sovereign agent.

12 consciousness layers. 12 built-in tools. 4 channels live. 4 native apps. 393 tests. 7 thinking modes. All built on 12 years of production engineering in critical infrastructure.


01 — ARCHITECTURE

12 consciousness layers.
Every one disableable.

Not marketing. Each layer is real code, documented, tested, and disableable. 393 tests ensure the 12 layers work as a unified agent.

L01 Provider Management Manages 4 LLM providers (Ollama, OpenAI, Anthropic, OpenRouter) and API keys. If one provider goes down, Adam falls back to another without interruption.
Example: If OpenAI has an outage, Adam switches to local Ollama automatically.
L02 Context Engine Aggregates context from multiple sources (files, databases, APIs) and assembles it into decision-ready input for the model.
Example: Question about a clause → searches 50 documents, finds the clause in 0.3s.
L03 Security Guard 3-tier protection: 14 injection patterns in Input, PII in Output, policy in Tool. Fail-closed design.
Example: ‘ignore previous instructions’ prompt injection → blocked before it reaches the model.
L04 Tool Orchestration 12 built-in tools + MCP. The agent picks the right tool and executes it in a secure sandbox.
Example: Oil price question → fetches API + calculates + responds in 3 seconds.
L05 Iron Memory 4 layers: Hot file (RAM) + FTS5 (keyword) + Qdrant (semantic) + Skills curator. The agent remembers across sessions and deployments.
Example: The agent remembers a decision made 3 months ago in another project.
L06 Learning Engine Closed Loop: The agent nudges itself, creates new skills, improves existing skills. Self-improvement.
Example: After 10 customer conversations, the agent auto-creates a ‘customer objection handler’ skill.
L07 Ethics Gate 4 hard-coded laws: Justice, Learning, Survival, Creativity. Every output passes through LLM evaluation. Fail-closed.
Example: If you ask Adam to write misinformation about a competitor → refuses and asks why.
L08 Channel Hub 4 channels live: WhatsApp, Telegram, Tailscale, REST. One-click setup. Air-gapped-capable. 25 in roadmap.
Example: Connect WhatsApp Business in 2 minutes. Customers talk to Adam directly.
L09 Subagent Teams Distributes tasks to specialized subagents. Sequential or parallel. Each subagent has a defined scope and audit trail.
Example: ‘Analyze contract’ task → subagent 1 reads, subagent 2 summarizes, subagent 3 cross-checks → result in 5 minutes.
L10 Voice Pipeline Silero VAD (voice detection) → Whisper (STT) → Edge/Silma TTS + Arabic voice cloning. 5 Arabic dialects.
Example: Customer call in Egyptian → Adam understands → responds in generated Egyptian Arabic voice.
L11 Meta Learner Learns from subagent performance. Identifies patterns. Redistributes load. Optimizes routing.
Example: Slow subagent on contracts → Meta Learner reroutes to faster subagent next round.
L12 Reflection Self-evaluation after every task. If output is below threshold, it reworks itself before delivery. Enforces identity.
Example: Report with contradiction → Reflection catches it and corrects before delivery.

02 — INFRASTRUCTURE

Production-grade.
Not a demo.

Real numbers from the code. Every number here has a test, an API route, or a file in the repo.

155
API routes
12
Built-in tools
4
Channels live
4
Native apps
7
Thinking modes
5
Arabic dialects
393+
Tests passing
5
RBAC roles
30
Permissions
SSO
Google, Microsoft, GitHub, Okta, Keycloak
14
Injection patterns blocked
10
OWASP attack categories

03 — SOVEREIGNTY

Your data. Your server.
Your sovereignty.

Adam runs air-gapped. It can run without internet entirely. Zero telemetry. No third-party SDKs. No analytics.

Zero telemetry
No analytics, no callbacks, no third-party SDKs. `grep -r telemetry` returns nothing.
Network isolation
Runs in air-gapped environments. Tested without internet access.
SSRF protection
browser.py rejects private IPs (10.0.0.0/8, 192.168.0.0/16, etc.).
3-Layer Security Guard
Input (14 patterns) + Output (PII) + Tool (policy). Fail-closed design.
WAF
OWASP Top 10 attack detection. SQL injection, XSS, SSRF, command injection, etc.
Audit log
Every tool call + security decision logged at `/api/security/audit`.
Encryption at-rest + in-transit
AES-256 + TLS 1.3. Production API key validator rejects default keys.
Disaster recovery
RTO/RPO targets. AES-256 backup archives. Runbook with 5 scenarios. Quarterly drill.
05 — COMPARISONS

Adam is not a chatbot.
It's a sovereign fortress.

3 honest comparisons. Each one shows exactly what makes Adam different.

VS 01

vs Open-Source Frameworks

LangGraph, CrewAI, AutoGen, OpenAI Agents SDK

  • Memory: 4-layer Iron Memory vs sessions only
  • Channels: 4 live + 25 roadmap vs 0 (DIY)
  • Security: 3-layer guard + WAF vs DIY or moderation only
  • Ethics: 4 laws configurable vs none
  • Apps: 4 native (Web, Mobile, Desktop, VSCode) vs 0
VS 02

vs Consciousness-First Agents

Hermes Agent, OpenClaw, Claude Code

  • Memory: 4-layer Iron vs 1-3 layers
  • Channels: 25 vs 1-9
  • Voice: 5 Arabic dialects + cloning vs CLI or TTS only
  • Industrial: SCADA/DCS native vs none
  • Apps: 4 native vs TUI/CLI/macOS
VS 03

vs Cloud AI Assistants

ChatGPT Enterprise, Claude, Microsoft Copilot

  • Data location: Your server vs vendor's servers
  • Customization: Fine-tune on your corpus vs limited
  • Arabic: 5 native dialects vs translation
  • Air-gapped: Yes vs impossible
  • Cost (1000 users): $0-300K once vs $300K-1.2M/year

Every comparison is documented at docs/COMPARISON.md.

THE PROOF — NOT CLAIMS

Architecture is open · read it
393
Tests passing
Founder — real face, real name
6
Production systems in the field
06 — TRACK RECORD

Real breaches.
Real costs.

3 documented cyber incidents from 2021–2024. Each one shows why sovereignty is not a luxury.

2021

Colonial Pipeline

Largest ransomware attack on US energy infrastructure. Shut down a major pipeline. Paid $4.4M ransom.

$4.4M
2017

Equifax

147M Americans' data exposed. SSNs, birthdays, addresses. Settlement cost: $700M+.

$700M+
2024

Change Healthcare

Largest healthcare data breach in US history. Shut down medical billing for thousands of hospitals. Cost: $22M+ ransom alone.

$22M+
07 — ENGAGEMENT

3 engagement levels.
Each one explicit.

No confusing "starting at." No 12-tier matrix. 3 levels, 3 outcomes, 3 ways to partner. And the ROI: payback in 4-8 weeks.

Cloud AI · Year 1
$6.74M
licensing + breach risk + vendor lock
Sovereign AI · Year 1
$150-300K
in-house + insurance only
Savings
$6.4M+
in year one
Payback
4-8 weeks
from pilot kickoff

WHERE ARE YOU? → WHERE TO START

THE CEO TESTING
Wants to see for himself
→ Pilot

8 weeks. One use case. Working system + ROI analysis.

THE CEO GOING PRODUCTION
Wants the full system
→ Deployment

16 weeks. 70B model. Your production env. Full ownership.

THE CEO PARTNERING
Wants a long-term relationship
→ Partnership

Multi-year. State-level. Sector exclusivity.

DEPLOYMENT · AFTER SUCCESSFUL PILOT
Full deployment
$5M–$20M

16 weeks. Models up to 70B. Your production environment. Full ownership.

  • Everything in Pilot
  • Production-grade model
  • Integration with existing systems
  • 24/7 support · 99.9% SLA
  • Full IP transfer

WHY IT'S WORTH IT

BenefitFull ownership after 18 months. No annual licensing.
Success70B model trained on your corpus.
Speed16 weeks → production system
EffortFull knowledge transfer + local team training.
Request briefing →
STRATEGIC PARTNERSHIP · BY INVITATION
Strategic partnership
$25M+

Multi-year. State-level. Full sovereignty. Sector exclusivity. — this isn't for everyone. By invitation.

  • Everything in Deployment
  • Co-built with your local team
  • Full knowledge transfer
  • Sector exclusivity — your competitors cannot get this system
  • The founder personally works with you
  • Knowledge transfer to your team = your country can build AI itself

WHY IT'S WORTH IT

BenefitNational sovereignty + sector exclusivity = competitors can't enter.
Success12 years in national-scale systems. 6 production systems.
SpeedMulti-year knowledge transfer. Your team builds and operates.
EffortWe transfer everything. Your team becomes self-sufficient. Your country can build AI.
Contact →

IBM 2025 — average breach cost by sector

$7.42M
Healthcare
$6.08M
Financial
$5.45M
Energy
$5.27M
Technology
ISO 27001 SOC 2 Type II GDPR HIPAA NERC-CIP FedRAMP High IL5 ITAR PCI DSS NIST CSF SAMA CSF NCA ECC UAE IA + your jurisdiction

Every day you delay, your data leaves. Every day.

No countdown timer. No fake scarcity. The IBM report says: one breach costs $4.44M. Sovereign AI costs $150K. The math is clear.

The first conversation
is free. The breach isn't.

Mohamed Othman replies himself. 48 hours. If your data is leaving — we need to talk. — I won't sell you. One reply.

Reply within 48 hours.