The software is CC-NC-BY licensed and free forever. Hardware BOMs are published. Training courses are open. Revenue comes from consulting engagements.

What hardware do I need for local AI?

Any modern desktop or laptop with 8+ GB RAM can run small models. For production, Edge tier starts at $1,269 with Framework Desktop + AMD GPU.

AINetGuard: Sovereign AI Infrastructure You Own

The Metal Layer

Hardware Infrastructure

We do not sell hardware, we help you build it. This is a reference for sovereign local-AI rigs you own outright, from compact edge desktops to multi-GPU rack clusters. Use the planner to scope a build, then bring it to us: we advise on parts, assembly, and setup so inference runs on your own metal with zero API calls and zero data export.

Open Hardware Planner & Calculators

Build planner, bill of materials by tier, inference-speed simulator, and GPU comparison — on their own page.

01 · Edge Tier In stock

Framework Desktop

All-in-one APU for local inference up to 70B quantized.

GPU8060S

MEM128 GB

BW256 GB/s

From $1,269 · Shop at frame.work ↗

02 · eGPU Tier In stock

RX 7900 XTX (eGPU)

External GPU + laptop. Thunderbolt 4 hot-plug sovereign inference.

GPU7900 XTX

VRAM24 GB

BW960 GB/s

From ~$1,050 · View on AMD ↗

03 · Research Tier Available

RTX 5090 (AM5 / TR)

Desktop or rack workstation. 1–10 GPUs for production inference.

GPURTX 5090

VRAM32 GB

FP4419 TFLOPS

From ~$1,900 · View on NVIDIA ↗

Published BOMs. No vendor lock-in. No hidden costs.

The Agentic Core

Software

40 open-source tools. No subscriptions, no API keys, no usage limits. Chat, art, voice, video, code, all local, all CC-NC-BY licensed. After setup: your team owns the same AI workflows that cost $200/month per seat on cloud platforms. Fork it. Modify it. Run it forever.

01 — In-House Software Agentic Core Features

Agentic Chat Interface

Local-first conversational AI with tool use, persistent memory, and human-in-the-loop gates. Every inference stays on your metal.

GitHub CI/CD Integration

Agentic code review, security scanning, and automated deployment plugged into your GitHub workflows. No vendor lock-in.

Multi-Agent Orchestration

Coordinate agent fleets with shared context, conflict resolution, and instant kill-switch authority. DAG-based execution for auditable workflows.

Deep RAG Engine

RAG anchored to your local data. Semantic chunking, hybrid search, and citation tracking, documents never leave your network.

02 — Recommended Stack Off-the-Shelf Software Catalog

Recommended Stack

Recommended · Not built by us

Off-the-Shelf Software

40 open-source tools that run on hardware you own. Chat, art, voice, video, code, security training, prompt engineering, all local, all free. After setup: a complete AI workbench without a single cloud dependency.

Loading tools…

03 — Interactive AI-Powered Demos

AI Tower Defense

3D tower defense powered by a live AI Commander. 15 waves, 5 tower types, 5 enemy types. Your AI advisor analyzes threats and outputs BUILD: and UPGRADE: commands you execute with one click. Click to play →

100% open source. CC-NC-BY licensed. Fork it, modify it, run it air-gapped. No telemetry, no license keys.Pick your hardware →

The Shield Wall

Security Tooling

Running AI locally is the first step toward sovereignty. Securing it is the second. Four layers of protection, from the model runtime to the data it touches, built on OWASP Secure-by-Design principles from day one.

01 — Runtime Security How to Secure Your Local LLM

The model runtime is your first attack surface. Every agent needs an identity, every action needs a permission check, and you need the ability to shut everything down in one click.

Agentic IAM (Zero-Trust)

DIDs for every agent. Granular permissions, automatic credential rotation, immutable audit trails, no centralized identity provider to compromise.

Runtime Isolation

Each LLM instance runs sandboxed with namespace isolation, restricted syscalls, and no outbound network unless explicitly allowed. Air-gapped by default.

Emergency: Terminate all active agents immediately

02 — Supply Chain How to Secure What You Load

Every model weight file, every dependency, every plugin is a potential supply chain attack. Verify before you trust, then harden what you deploy.

Automated Penetration Probing

Continuous offensive testing. Prompt injection, privilege escalation, data exfiltration, found before real attackers find them.

Code Hardening (Defensive)

Automated hardening of agent outputs. Input validation, output sanitization, and dependency scanning in your CI/CD pipeline.

Model Provenance & Integrity

SHA-256 hash verification for every GGUF weight file. Track model lineage from source repo to local disk. Reject unsigned or tampered weights automatically.

03 — Data Sovereignty How to Secure Your Data

Your prompts, your documents, your conversations, they never leave your metal. Real-time monitoring catches anomalies before they become breaches.

Semantic Drift & Cost Observability

Track agent behavior drift over time. Real-time dashboards for token cost, resource consumption, and anomaly detection.

Zero-Export Data Policy

Network-level enforcement: no telemetry, no usage reporting, no API callbacks. Your inference data stays on your hardware, verified by firewall rules you control.

Active Agents

Semantic Drift

0.02%

Token Cost / hr

$4.82

Anomalies (24h)

04 — Roadmap What’s Next

Security doesn’t stand still. These capabilities are in active development, battle-tested in our CTF sandboxes before they ship to production.

Localized CTF Sandboxes

Local capture-the-flag environments for stress-testing agent security. Run adversarial scenarios without exposing production. Learn AI →

Autonomous Agent Governance

Formal constraint frameworks for multi-agent systems. Define what agents can decide, what requires human approval, and what is never allowed, enforced at the runtime level.

The Human Layer

Consulting

Pentest your AI. Harden your apps. Align with regulators. Respond to incidents with a 4-hour SLA. Most teams deploy AI without knowing what they don't know. After engagement: your team owns merge-ready fixes, a compliance-ready audit trail, and a playbook for the next incident.

AI Pentesting & Pentesting AI

Our adversarial AI agents test your AI agents in safe, controlled environments. Prompt injection, jailbreaking, data exfiltration, privilege escalation, we find the gaps together so you can close them with confidence.

Web App & API Pentesting

Traditional application security testing for every surface your agents expose to the world. OWASP methodology, rigorous manual testing, and actionable reporting with severity-ranked findings.

AppSec Remediation

We don't just find vulnerabilities, we help you fix them. Code-level remediation, architecture redesign, and infrastructure hardening, all delivered as merge-ready pull requests your team can review and learn from.

Autonomous Liability Strategy

Legal and technical frameworks for agentic liability. Who is accountable when an agent acts autonomously? We'll help you build clear answers before regulators come asking, no surprises, just preparedness.

Regulatory Alignment

EU AI Act, NIST AI RMF, ISO 42001, navigating these frameworks can feel overwhelming. We'll map your agentic infrastructure to every relevant requirement and build the compliance documentation alongside your team.

Rapid Incident Response

Active breach containment, agentic failure forensics, and guided recovery. We're available within 4 hours and stay with you through the entire process, including comprehensive post-incident reporting.

Free first consultation on every engagement. No sales calls, just engineers.Hardware tiers → | Learn AI →

The Knowledge Layer

Learn AI

Most kids use AI unsupervised, on cloud platforms that collect their data and shape their thinking. From your child's first prompt to your security team's OWASP certification, all on hardware you own. After completion: your family and your team both understand AI, control it, and own every piece of the stack. AI literacy is civic infrastructure.

🧠 Kids 9–13 🎓 Professional

🧠 Learn to Prompt — Ages 9–13

Project-based prompt engineering. On your hardware. With parent visibility.

Think Before You Prompt

How AI actually works, not magic, not sentient. What it’s good at, what it’s bad at, and why the way you ask matters more than what you ask.

Write Clear Prompts

Structure, specificity, examples, and constraints. The same skills that make good writing make good prompts, and your kid learns both.

Check the Output

AI makes things up. Confidently. Kids learn to fact-check, spot hallucinations, and never trust an answer they can’t verify.

Stay Safe & Private

What data AI collects, why local matters, and how to use AI without giving away personal information. Practical digital literacy.

Build Real Projects

Not worksheets, real projects. A story generator, a homework helper, a quiz maker, a code buddy. Each one teaches a prompt engineering concept.

Parent Dashboard

See what your child asks, what the AI answers, and how they’re progressing. No surveillance, just visibility. Running on hardware in your home.

🎓 Professional Training — Free Courses

No credit card, no registration wall. Our free tier is not a marketing funnel. It is the product.

Free

OWASP Top 10 for AI

The ten most critical AI security risks. Agentic attack vectors, practical mitigations, and case studies from real production incidents.

Free

Hands-on CTF for Developers

Capture-the-flag exercises targeting real AI/ML vulnerabilities: prompt injection, model poisoning, data exfiltration. Runs locally.

Free

AI Security Fundamentals

Threat modeling for AI systems, secure deployment patterns, and the security implications of autonomous operation.

Pro

Enterprise Phishing & Security Culture

AI-powered phishing simulation, culture-based defenses, and measurable security behavior change programs.

Pro

C-Suite AI Adoption Strategy

Board-ready frameworks for AI risk, ROI modeling, and governance. Tailored to your industry vertical.

Pro

Agentic Systems Architecture

Design, deploy, and govern multi-agent systems. Tool-use orchestration, guardrails, observability, and failure containment.

Bring live AI training to your team

A tiered, instructor-led AI training program delivered in person at your location. An intensive bootcamp that takes a whole team from "what is AI" to security-specialist depth, with safety engineered into every level. 6 courses. 36 modules. A live trainer, your people, your site.

On-Site Delivery Instructor-Led Team Training Safety built in

Entry Level AI Foundations

11 modules 🛡 Safety by Design built in

Part 1: Prompting & Workflow Fundamentals

1.1 Prompting Fundamentals
What's covered
- What a prompt actually is and how phrasing changes output
- Anatomy of a strong prompt: role, context, task, format/length
- Common failure modes and quick fixes
- Taxonomy of Prompting
1.2 Context Window & Data Management
What's covered
- What context is and why the model only knows what's in front of it
- The context window: limits, overflow, managing long inputs
- Organizing and feeding data (chunking, summarizing, ordering)
- When to start fresh vs. continue a thread
1.3 Agentic Workflows: Executing Work with Cloud AI
What's covered
- What 'agentic' means in plain terms
- Using cloud AI to execute a multi-step workflow
- Example workflows relevant to the audience
- Where a human checkpoint belongs in the loop
1.4 AI User Security & Safety
What's covered
- Dangerous Data
- Dangerous Activity
- Secure Antigenic Usage
1.5 Safety by Design: Rules, Guards & Safe HabitsFlagship · Safety
What's covered
- Safety by design: setting yourself up so mistakes can't cause harm
- Acceptable-use rules and simple personal guardrails
- What never to share; recognizing when to stop and escalate
- Building verification into your everyday routine
1.6 Research Pipeline
What's covered
- A repeatable flow for researching a topic end-to-end
- Sourcing, cross-checking, and citing
- Synthesizing findings into a usable output
- Guarding against hallucinated sources

Part 2: Tech Fundamentals

2.1 Managing Cloud AI
What's covered
- What 'cloud AI' means and how it's accessed/managed
- Accounts, access, and basic administration
- Cost / usage awareness
2.2 Connecting a Codebase or Documents
What's covered
- Ways to connect AI to your own material (docs, repos, KBs)
- What 'good' connected context looks like
- Practical setup walkthrough
2.3 Priming AI for Different Teams & Tech-Training Needs
What's covered
- 'Prime techniques': different ways of running AI for different jobs
- What different departments need + the tech training each requires
- CAUTION: don't put env / secrets into a local AI (anti-patterns to avoid)
2.4 Where AI Lives: Business Impact & Data Residency
What's covered
- Where AI runs (local vs cloud) and where data physically lives
- Business impact of that choice
- Data residency / handling (operational; legal is 2.5)
2.5 Legal Considerations: IP, Local vs Cloud
What's covered
- Intellectual-property issues with AI use
- How local vs cloud changes the IP / ownership picture
- Practical guardrails and 'ask legal when…' triggers

Why on-site training

Built and delivered by a sovereign-AI security company. A live instructor at your site, not a video course.

📍

We come to you

A live instructor delivers the training on your premises, using your team, your environment, and your tools. No travel for your people, no scheduling around a public cohort.

👥

Built for teams

Train a whole team together, at one depth or across tiers. Shared context, shared exercises, and a common vocabulary your people keep using after we leave.

🛡

Safety by design

Every Entry and Intermediate course has a flagship Safety-by-Design module plus a safety thread woven through every other module. Not a bolt-on.

🎯

Tailored to your stack

We tune the curriculum to your team's size, background, and tools. Pick a tier, choose a specialization, and we shape the on-site sessions to fit.

Questions people ask

Is this actually free, or is there a catch?

No catch. The software is free and always will be. We publish exactly what hardware to buy and where to get it. The training courses are open to everyone. We make money when companies hire us for hands-on help, not by locking you into a subscription or harvesting your data. The free stuff isn't bait. It's the whole point.

What hardware do I need to get started?

Any modern desktop or laptop with 8+ GB RAM can run small models via Ollama. For production workloads, our Edge tier starts at $1,269 with a Framework Desktop + AMD GPU. The hardware calculator in our Hardware section will tell you exactly what fits your use case.

How is this different from using ChatGPT or Claude?

Cloud AI sends every prompt to someone else’s servers. Local AI runs on your hardware, your data never leaves your network. You control the model, the context window, the guardrails, and the kill switch. No rate limits, no API costs, no vendor lock-in.

Can I use this for my business / in production?

Yes. Everything is CC-NC-BY licensed. For commercial deployments, our consulting team handles hardware specification, security hardening, agentic architecture, and ongoing support. Every engagement starts with a free consultation.

Is this safe for my kids to use?

That’s exactly what Learn to Prompt is for. We teach kids 9–13 to use AI on your hardware with your visibility. No cloud accounts, no data collection, no unsupervised access. The AI runs locally, you can see every prompt and every response.

Train your team on AI, at your site

Hardware Planner & Calculators

AI Apps

Agentic Pentesting

Safe Links

Local AI & Models

AI & Application Security

Reference Material

The AINetGuard Manifesto

Entry Level

Intermediate

Advanced

Entry Level

Part 1: Prompting & Workflow Fundamentals

Part 2: Tech Fundamentals

Intermediate

Core Track: Local Model Mastery

Bonus: Security & Safe UseBonus

Track: Agentic & Personal AI Engineering

Security In Depth (Attack & Defense)

Advanced

Track: AI-Augmented Software Engineering

Track: AI Infrastructure & Deployment

Track: Offensive Security & Forensics

Why on-site training

We come to you

Built for teams

Safety by design

Tailored to your stack

Questions people ask