We set up AI that actually works for your life.
Without leaking your medical practice into someone's training set. Without your family sharing one account with no governance. Without browser extensions you installed once and forgot about storing everything you've ever typed. Sentinel configures AI the way it should have been set up the day you subscribed.
None of this is carelessness. It's what happens when a consumer product gets handed to a household with no configuration, no governance, and no one responsible for the outcome.
You pasted a client contract into ChatGPT to get a summary. Your associate pasted a patient intake form into Claude to draft a letter. Your CFO ran a revenue projection through Gemini to format a table. Consumer accounts on default settings contribute conversational data to model training pipelines — the terms are clear, the behavior is expected, and the data is gone. A physician who pastes patient records into a consumer AI has created a HIPAA exposure event. A real estate principal who pastes deal terms into a consumer chatbot has potentially disclosed confidential transaction details. The tool works. The configuration is wrong.
Your ChatGPT account has your billing information, your custom instructions ("I am a physician managing a practice of 40 patients..."), your full conversation history, and your uploaded documents. Your teenager uses the same account to do homework. Your housekeeper uses it to draft messages. Your spouse uses it to plan travel. Every conversation, every uploaded document, and every configured system prompt is visible to everyone logged in. There are no separate roles, no access controls, no audit trail. A shared AI account is a shared context window — everything anyone typed is context for everything anyone else does next.
You have Claude Pro, ChatGPT Plus, Gemini Advanced, and Perplexity. You open whichever tab happens to be in your browser. Different models have different strengths, different context windows, different privacy configurations, different data retention policies, and different performance profiles on different tasks. Using the wrong model for sensitive legal drafting is a privacy risk. Using the wrong model for research is a quality risk. Using the wrong model for code is a correctness risk. Nobody in your household knows which tool to use for which job. There is no policy, no training, no documentation. The subscriptions cost $200/month and the output quality is random.
You installed a Chrome extension that "makes ChatGPT better." Your assistant installed an AI email plugin. Someone on your team enabled a Zapier integration that reads your email to auto-summarize threads. These integrations have broad permissions: read your browser history, access your email, read your clipboard, inject content into every webpage you visit. Most are legitimate products built by small teams with no enterprise security review, minimal data retention policies, and no breach notification program. An AI productivity tool installed without review is an open channel from your browser to someone else's server. Most clients have four to six of them active and can name none of them from memory.
One engagement. Every account configured, every integration audited, every family member trained on which tool does what. You stop guessing and start using these tools the way they were designed for someone with something to protect.
Every AI subscription you use for anything work-adjacent moves to a Team or Enterprise account. ChatGPT Team, Claude for Work, Gemini for Workspace — these accounts have data-off-by-default settings, conversation data that doesn't feed training pipelines, and per-user access controls so your household and your practice are separated. We migrate your existing personal subscriptions, configure the data controls correctly, and verify each account's current data retention posture against its published privacy policy before sign-off.
For clients with practices, firms, or funds that handle regulated or confidential data, we configure private LLM deployments: Azure OpenAI Service (HIPAA BAA available, data stays in your Azure tenant), AWS Bedrock (regional data residency, enterprise access controls), or local Ollama deployments for the most sensitive workflows (no network call, runs on hardware you own, zero data leaves the device). We scope the right option to the sensitivity level of the work, not the maximum configuration by default — local Ollama for a physician drafting referral letters is the right answer; Azure OpenAI for a family office doing deal research is the right answer.
We build a private knowledge base over your documents: estate planning documents, investment memos, property records, operational procedures. You ask questions in plain English, the system retrieves relevant context from your own files, and the model answers from your documents rather than from the internet. The documents never leave your infrastructure. We scope, build, and deploy this using a private vector store (Pinecone or Chroma, self-hosted) and a secure API layer you control. When you ask "What does my trust say about real property held in Nevada?" you get a sourced answer from your actual documents, not a hallucinated response.
Every voice assistant in your home is configured for minimal data retention: review history disabled, voice recordings deleted on a rolling schedule, purchase history access removed, third-party Alexa Skills audited and unused skills revoked. Siri is configured with Private Cloud Compute enabled so voice queries process on-device or in Apple's private relay infrastructure rather than on shared servers. We go room by room, device by device, and leave you with a written record of every setting changed and every account permission removed.
Apple Intelligence routes requests through Private Cloud Compute — Apple's attestation-based infrastructure where requests are processed on Apple Silicon nodes with no persistent storage and no Apple employee access. We verify Private Cloud Compute is enabled on every Apple device in your household, confirm which AI features are on-device versus cloud-routed, and document the data flow for every AI feature you use. We configure Siri's Personal Context access (what Siri can see across your apps) appropriately for each family member's role. Your phone assistant knows what you need it to know. Nothing more.
This is what Sentinel builds during an AI setup engagement. Not consulting recommendations. Not a vendor evaluation deck. Configured, tested, handed to you working.
Individual accounts per family member — adults and children separately, with children's accounts configured under supervised access. Every account has training data off, conversation history managed, and appropriate system prompts configured so the tool behaves correctly for the person using it. Shared documents and context live in Team vaults, not personal history. Each family member gets a 30-minute walkthrough of their setup before we leave.
We ingest the documents you want to query — estate documents, investment records, property files, operating agreements — into a private RAG system you own and control. You ask questions. The system returns sourced answers from your files. Nothing leaves your infrastructure. We document the ingestion pipeline, the query interface, and the maintenance procedure so you can add documents yourself after handoff.
For high-volume inboxes, we configure an AI triage layer that categorizes, flags urgency, and surfaces action items — without reading or storing your email content on third-party servers. The integration uses your own AI account with appropriate scope permissions and runs within your existing email client. You see a summary of what needs your attention. The AI sees only what you authorize it to see, within the data retention policy of the account you already control.
We deploy a meeting transcription and summary workflow — Granola or a self-hosted alternative — configured with explicit retention rules: transcripts deleted after 30 days by default, action items exported to your task manager, no data shared with third-party services without explicit per-meeting authorization. Client meetings and sensitive conversations are excluded from AI note-taking by default. The workflow is documented and every team member who uses it is briefed on what the tool does and does not retain.
For clients who want AI-powered automation (drafting, routing, summarizing, scheduling), we build playbooks in Zapier or Make with a full audit trail: every automation logged, every data source documented, every third-party service reviewed for its data handling policy before connection. You get working automations that save real time. You also know exactly what is automated, what data each automation accesses, and how to shut any of them off. No black boxes running unreviewed in your name.
Consumer AI advice stops at "use the enterprise tier." Sentinel goes further — configuring compliant workflows, documenting your data posture in plain English, and making sure the tools your team uses daily are not creating regulatory exposure you haven't thought about.
We configure AI tools for medical practices under a HIPAA-compatible posture: Azure OpenAI Service with a Business Associate Agreement, on-device summarization for anything touching PHI, intake and note workflows that process patient data inside your HIPAA-covered environment. We do not configure consumer ChatGPT or Claude accounts for PHI workflows — we tell you that clearly and build the right alternative instead. Your AI tools work. Your compliance posture is documented. Your Business Associate Agreements are in place before go-live.
We configure AI tools for deal workflows — lease abstraction, due diligence memo summarization, comparables analysis — within a private deployment that keeps deal documents off shared infrastructure. We audit every integration your team uses to confirm no deal data is flowing through consumer AI accounts. We configure data-off settings on every account that touches confidential transaction information. The result is a team that uses AI productively in the deal room, with a documented chain of custody for every sensitive document that enters the workflow.
We configure private research pipelines — RAG over your own research archive, AI-assisted memo drafting within your own infrastructure, competitive intelligence workflows that pull public data without exposing proprietary analysis. For family offices, we configure the principal's AI workflow to be separated from the household AI configuration: different accounts, different data retention policies, different system prompts. Portfolio analysis and estate planning documents do not live in the same context window as your teenager's homework assistant.
At the end of every AI setup engagement, you receive a one-page compliance posture document: what AI tools your household and practice use, what data each tool accesses, what the current data retention policy is, what your exposure would be in a breach event, and what the next review date should be. Not a legal opinion. Not a compliance audit. A plain-English record that you can hand to your attorney, your HIPAA privacy officer, or your family office compliance consultant as a starting point. Most clients have never had this document. It takes us a day to produce it.
Every month you run on default AI account settings is a month of conversation history building up in accounts you've never reviewed. Book a discovery call or request an assessment — AI setup is included in every Sentinel engagement and scoped to your specific household and practice.
Related: Shared-account hygiene → · AI account inheritance in your digital estate →