Private AI: Deploying an On-Premises Assistant for Your Organization
For IT and platform teams: plan, build and operate a self-hosted AI assistant on your own hardware, with open-weight models, a chat front end, single sign-on, document search and governance your security team will accept.
The curriculum is being prepared and reviewed. Enrollment is not open yet.
About this course
Many organizations cannot or will not send sensitive data to a public AI service, yet their staff want an assistant as capable as the consumer tools. This course shows how to deliver one on hardware you control. You begin with requirements, the data classes involved and the honest comparison with private cloud options, then size hardware: GPU memory for weights and KV cache, concurrency, and the trade-offs between data-center GPUs, workstation-class cards such as the NVIDIA RTX PRO 6000 Blackwell, and small appliances.
You select models by license, capability and size, serve them with vLLM (with Ollama for evaluation), and deploy a front end: Open WebUI or LibreChat, including Open WebUI's branding clause for deployments above 50 users. You integrate identity through OIDC with Microsoft Entra ID or Keycloak, map groups to models and tools, connect document search over SharePoint and file shares with permission-aware retrieval, and add tool and MCP integrations carefully. The final modules cover air-gapped installs, model and container supply chain, logging and retention, acceptable-use policy, capacity monitoring and the upgrade cycle. Single-user local tools are covered in ven-290 and serving internals in srv-310. Facts are stated as of October 2026. The course ends in an automated exam; passing earns a Wirewalk Institute of AI certificate of completion and assessment.
What you will be able to do
Write requirements for a private assistant covering data classes, users, concurrency and residency
Size GPU hardware for a target model, context length and concurrent user count
Select open-weight models by license, capability, language coverage and memory footprint
Deploy vLLM behind Open WebUI or LibreChat with OIDC single sign-on and group-based access
Connect internal documents for retrieval while enforcing source-system permissions
Operate the service with monitoring, logging and retention policy, model updates and an air-gapped supply chain
Present total cost of ownership and a support model to leadership
Syllabus
6 modules · 20 lessons · about 14 hours
1
Module 1: Requirements and the Case for Private AI
Deciding what must stay on premises and what the service must do.3 lessons
Why On-Premises: Confidentiality, Residency, Air Gaps and Cost Control
Requirements Workshop: Users, Use Cases, Data Classes and Concurrency
On-Premises, Private Cloud or Enterprise SaaS: An Honest Comparison
2
Module 2: Hardware and Model Selection
Matching models to GPUs and users.4 lessons
Sizing GPU Memory: Weights, Quantization, KV Cache and Concurrent Users
Hardware Options: Data-Center GPUs, RTX PRO 6000 Blackwell Workstations and Small Appliances
Choosing Open-Weight Models: gpt-oss, Qwen, Mistral, Gemma and Llama Compared
Model Licenses and Acceptable-Use Terms for Internal Deployment
3
Module 3: The Serving and Application Stack
From model server to the chat interface staff actually use.3 lessons
Serving with vLLM: OpenAI-Compatible API, Multiple Models and Tool Calling
Open WebUI Versus LibreChat: Features, Licensing and Operations
Deploying with Docker Compose or Kubernetes, TLS and Reverse Proxies
4
Module 4: Identity, Access and Integrations
Making the assistant part of the organization's systems safely.3 lessons
Single Sign-On with OIDC: Microsoft Entra ID and Keycloak
Groups, Roles and Per-Model Access Control
Tools, Functions and MCP Servers: Useful Integrations Without Excess Privilege
5
Module 5: Searching Internal Documents
Grounding answers in company knowledge without leaking it across teams.3 lessons
Connecting SharePoint, File Shares and Wikis
Permission-Aware Retrieval and Document-Level Security
Local Embedding Models, Vector Stores and Re-Indexing
6
Module 6: Operating and Governing the Service
Running a private assistant as a supported, trusted service.4 lessons
Air-Gapped Installation and Model and Container Supply Chain
Monitoring, Capacity Planning and Logging with Retention Limits
Acceptable-Use Policy, Training and User Support
Total Cost of Ownership and the Model Upgrade Cycle
Exam and certificate
The exam unlocks once you have completed every lesson. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 40 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.
Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.
The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by Open WebUI, LibreChat, vLLM, Ollama, NVIDIA, OpenAI, Alibaba Cloud, Mistral AI, Google, Meta, Microsoft, or Keycloak. Product and company names are trademarks of their respective owners.
A hands-on course on building no-code assistants with the builders already in your workplace tools: ChatGPT projects, custom GPTs and their successor plugins, Claude Projects and Skills, Gemini Gems, and Microsoft 365 Copilot Agent Builder and Copilot Studio. You leave with a tested, documented assistant and a plan to share it safely.
A hands-on, code-first course on the three major LLM APIs side by side: messages, streaming, structured outputs, tool calling, vision, cost, caching, batch, retries, abstraction layers, cloud routes and testing. You finish by building and hardening a small multi-vendor service.
A practitioner's course on retrieval-augmented generation: when to use it, how to ingest, chunk, embed and index documents, how to combine BM25 and vector search with reranking, and how to measure, debug and operate the result. Every module includes working Python sketches.