Understanding AI,
Claude Code & HAUTE
A Vituity Guide to Agentic AI
HAUTE Platform 2026
A Vituity Guide to Agentic AI
HAUTE Platform 2026Your journey from chatbots to agentic AI
Non-technical professionals who want to understand how AI, Claude Code, and the HAUTE platform work together to automate workflows, manage information, and boost productivity. No coding experience required.
You've almost certainly used some form of AI already — whether it's ChatGPT, Via (Vituity's internal ChatGPT tool), Copilot, or another assistant. All of these fall into the category of a chatbot.
A chatbot is a text-in, text-out interface. You type a question, you get an answer. You paste a document, you get a summary. Maybe you upload a photo and get a description. But the key limitation is that you're constrained within the chat session. The AI can only respond — it can't reach out and do things in your world.
| Layer | What It Is | Example |
|---|---|---|
| AI Company | The organization that builds the underlying AI models (called LLMs — Large Language Models) | Anthropic (makes Claude), OpenAI (makes ChatGPT) |
| Chatbot | A conversational interface for Q&A, writing, and analysis | Claude.ai, ChatGPT, Via |
| Agentic AI | AI that can take action — read your email, manage your calendar, write code, build things | Claude Code |
| Enterprise Platform | A layer built on top of agentic AI that adds memory, enterprise integrations, and guardrails | HAUTE |
Anthropic is the company that makes the LLMs. Claude Code is their agentic AI tool. HAUTE is the ecosystem Jeff McDonald built on top of Claude Code to make it enterprise-ready for Vituity.
The word "agentic" comes from agency — the ability to act. When you use a chatbot, you're having a conversation. When you use agentic AI, you're delegating real work.
Claude Code is Anthropic's agentic AI tool. Once set up through HAUTE, it has access to your entire Microsoft 365 environment and can perform actions across all of these systems:
Read, search, draft replies, and compose new messages. Creates drafts for your review before sending.
View your schedule, find available times, create meetings, and manage invitations.
Read messages, send replies, check presence, and participate in conversations on your behalf.
Search documents, read files, access shared content across the organization.
Access and analyze recordings from your Teams meetings, extract action items and summaries.
Read and write files, install software, configure settings, and build tools directly on your machine.
Instead of telling you how to do something, agentic AI actually does it. Here's a concrete comparison:
Agentic AI always asks for your permission before taking significant actions. With email, it creates drafts for your review rather than sending directly. You remain in control.
Claude Code was originally built to help software developers write programs. But here's the thing: none of us need to be developers. You have one of the world's smartest AIs available to do the coding for you.
Claude can create scripts, automations, dashboards, and even full applications. Through the course of your work, you almost certainly have manual, repetitive tasks that would benefit from automation. Claude Code makes that possible without writing a single line of code yourself.
Your job is to specify what you want. The better you describe it, the better Claude can build it. There are two general approaches:
Once Claude builds something useful, you can save it as a skill — a reusable automation you can invoke anytime. Preparing for meetings, pulling reports, analyzing data, triaging emails — each becomes a one-command operation tailored to your specific workflow.
For Claude Code to do useful work in your world, it needs to connect to your email, your calendar, your Teams, your knowledge base, and dozens of other services. The technology that makes this possible is called MCP — the Model Context Protocol.
An MCP server is a small adapter that knows how to talk to one specific service (like Outlook, or JIRA, or SharePoint). Claude Code speaks the standard MCP language. As long as a service has an MCP server, Claude can use it — no custom integration code required.
| MCP Server | What It Connects To |
|---|---|
| ht | The unified HAUTE CLI — M365 (email, calendar, Teams, SharePoint), JIRA, Confluence, ServiceNow, Neo4j memory, Obsidian vault |
| Web search | Brave, Exa, Tavily, Perplexity for live internet research |
| Gemini, OpenAI | Other AI models for image generation, reasoning, and transcription |
| Domain-specific | Additional MCP servers can be added based on your role (e.g., security tools for InfoSec) |
You don't need to think about MCP day-to-day — it just works. But understanding it helps explain how Claude Code can read your email, search SharePoint, look up a JIRA ticket, and check threat intelligence all in one conversation. Each connection is an MCP server, and HAUTE bundles dozens of them so you can plug in whatever your role needs.
HAUTE (Human-Augmented Unified Team Environment) is the ecosystem built on top of Claude Code that transforms it from a developer tool into an enterprise-grade AI platform for Vituity.
Your AI remembers across sessions — your preferences, your work, your context. It gets smarter over time.
Connects to Microsoft 365, JIRA, ServiceNow, Confluence, and more — all from one interface.
Hooks that block dangerous actions, enforce draft-only email, and prevent accidental deletions.
Pre-built capabilities for email, calendar, project management, legal review, coding, and more.
A unified command-line tool that connects Claude to email, calendar, Teams, memory, JIRA, and SharePoint.
Your HAUTE becomes uniquely yours — not an off-the-shelf generic AI, but one that knows you and your work.
Think of HAUTE as a big box of Legos. The platform provides all the building blocks — connectors, memory, skills, safety rails — and you assemble them into whatever you need. Your use cases will be unique to your role, and that's exactly the point: you build something custom, not generic.
Over 220 Vituity employees are now on HAUTE, including a dedicated clinical users group of 15+ clinicians. You can learn from peers, share skills, and adapt automations others have built to your own workflow.
If you've used ChatGPT or other chatbots, you've hit this problem: you have a great conversation, come back the next day (or even the next hour), and the AI knows nothing about what you just discussed. You're literally starting over every time.
HAUTE solves this with two complementary memory systems that make your AI genuinely personal and persistent:
Think of it like a reference library for your AI. Obsidian stores full documents, notes, files, and text blocks. You populate it by telling Claude to save things, dragging files in, or through automated actions the HAUTE system performs on its own. Over time, this becomes a rich knowledge base of your work, your context, and your domain expertise.
Organized with PARA:
Projects (active work with deadlines)
Areas (ongoing responsibilities)
Resources (reference material)
Archives (completed work)
A graph database that stores information as connected nodes. For example: "Ray" is a node with attributes (email, location, title). "IT Department" is another node. The connection between them ("Ray is a member of IT Department") is what makes the graph powerful. When your AI retrieves information, it traverses these connections like a chain of thought, linking related pieces of knowledge together.
You don't need to build this all at once. Memory accumulates naturally:
Memory transforms your AI from an off-the-shelf tool that anyone could use into a personalized assistant that understands your role, your projects, and your work patterns. It's the difference between a stranger and a trusted colleague.
Agentic AI is powerful, which means it needs strong guardrails. HAUTE is built with multiple layers of protection to keep you in control at all times.
A hook is a script that triggers automatically when certain actions occur. Think of them as guardrails that prevent mistakes before they happen:
| Hook | What It Does |
|---|---|
| Deletion Blocker | Prevents Claude from deleting files on your system. If it tries, a red warning appears and the action is blocked. |
| Draft-Only Email | Email compositions create drafts for your review, never sending directly. You always get the final say. |
| Permission Prompts | Before making any change, Claude asks for your approval. You say yes, no, or redirect. |
| Session Bootstrap | Loads your identity, memory, and preferences at the start of every session. |
HAUTE does not use the public Anthropic API. Instead, Vituity runs private Claude models in our own Azure AI Foundry instance. This is the same approach Via uses with ChatGPT — keeping data within Vituity's controlled cloud environment.
You won't always get 100% identical results between users or between sessions. AI uses probability to generate responses, so minor variations are normal. Always review important outputs before acting on them.
Claude Code can connect to external services like web search. Just as you wouldn't paste confidential information or PHI into a Google search, don't direct Claude to share sensitive data externally. The private models protect AI processing, but outbound connections to the internet are not contained.
HAUTE is set up through a guided session with the IT team. The installer configures everything for you, including Claude Code, the HT CLI, memory systems (Obsidian and Neo4j), enterprise connections, hooks, and skills. The process takes about 60–90 minutes.
| Component | Purpose |
|---|---|
| Claude Code | The agentic AI engine — your primary interface for all AI interactions |
| HT CLI | The HAUTE command-line tool for email, calendar, Teams, memory, and more |
| Obsidian | Your Digital Twin — a knowledge vault that grows with your work |
| Neo4j | Your Knowledge Graph — a database of connected information |
| Settings & Hooks | Safety guardrails, permissions, and enterprise configuration |
| 125+ Skills | Pre-built capabilities for common tasks |
Instead of "find that email," say "find the email from Sarah about the Q3 budget from last Tuesday." More context = better results.
Ask peers to share their skills. You can adapt what others have built to your own workflow in minutes.
Your first request won't be perfect. Refine it. Claude learns from the conversation and each iteration gets closer to what you want.