Overview Our Story For Family Offices Product Features Hardware Specs Privacy by Design Client Privacy Request Info
For Family Offices

The intelligence your office runs on.
Sovereign. Always on. Never exposed.

The average operating cost to run a single-family office above one billion dollars reached $6.6 million in 2026, according to J.P. Morgan's Global Family Office Report. A significant share of that cost is driven by the complexity of managing sensitive information that cannot safely touch a public cloud. Staff search manually. Documents are siloed. The AI tools the rest of the world uses freely are, for this office, a liability.

In June 2026, Citi and Fortune identified a second problem: a small but growing number of family offices are already exposed to AI through tools they did not knowingly adopt. SaaS platforms, daily-use devices, and productivity software have quietly routed sensitive data through cloud AI models. The back door was open before anyone thought to close it. The Obsidian Sovereign closes it.

A Conversation That Never Left Your Building Cannot Be Produced

A sealed local AI system has no third-party custodian to compel. In February 2026, a federal court in United States v. Heppner (No. 25-cr-00503-JSR, S.D.N.Y.) ruled that a defendant's exchanges with a consumer AI platform were not privileged under attorney-client or work product doctrine. Prosecutors used those conversations as evidence at trial. The lesson is plain: a cloud AI conversation is a discoverable record held by someone else. Every conversation your staff has on the Obsidian Sovereign stays inside your walls, encrypted, held by no one but you. There is nothing to subpoena that we hold. Nothing to produce. Nothing to compel.

Every Document Your Office Has Ever Touched, Instantly Searchable

Every document the office has ever created or received, contracts, correspondence, financial records, legal briefs, trust documents, investment memos, tax filings, is ingested into the Obsidian Sovereign and made instantly searchable by staff in plain English. When the principal calls and asks for facts and figures, the answer is always there. Staff can ask directly: what were the terms of the Morrison trust amendment from two years ago? The answer surfaces in seconds from documents stored on-premises, processed locally, with nothing transmitted to any external service. That is not a search feature. That is a private Google for your entire document universe.

Every Conversation Your Staff Has Belongs to You, Not a Provider

Every conversation your staff has with the Obsidian Sovereign returns to your own server. Searchable by your own AI. Across one external model or all of them, at your direction. The memory your office builds is yours, not a provider's asset, not a training dataset, not a record someone else holds. No cloud subscription works this way. This one does.

When You Choose to Reach Out, The Veil Protects You

The Obsidian Sovereign is sealed by default. Nothing leaves the building. When a staff member deliberately chooses to query an external AI platform, every prompt passes through the Veil, Apex Fortress AI's proprietary identity protection layer, before it touches the open internet. Names, entities, financial details, and identifying context are stripped and replaced. The external AI receives a clean query. The answer returns and is re-contextualized locally. The outside service never knows who asked, where the query originated, or what office it came from.

Always On. No Provider Can Take It Away.

Cloud AI is only as available as the provider's infrastructure. Outages, throttling, and service degradation are facts of life with every major platform. In 2026, a U.S. government export directive forced a major AI provider to globally disable two frontier models with no advance notice to clients. A sealed local system has no provider to receive such a directive. The Obsidian Sovereign runs on hardware inside your office. As long as there is power in the building, your staff has full access to the AI and to the entire ingested document corpus.

Intelligence That Runs Models Most Organizations Cannot Even Load

The Obsidian Sovereign runs on AMD's Strix Halo platform with 128GB of unified memory, hardware purpose-built for AI inference. It runs Llama 4 Scout, a 109 billion parameter model, at approximately 18 tokens per second. An RTX 4090, the GPU most organizations consider top-tier AI hardware, physically cannot load this model. The gap is not performance. It is capability. Your office runs AI that the hardware most organizations rely on simply cannot access.

Three Levels of Control. One Decision Each.

The Obsidian Sovereign includes three independent levels of failsafe control. The first seals the system instantly, locking the AI layer behind biometric authentication until you choose to reopen it. The second, on your command or by conditions you define, destroys the cryptographic keys that protect the local corpus. Without those keys, the data is unreadable by anyone, including us. The third wipes the encrypted Swiss off-site backup remotely. You hold the only key to that vault. Each level is a single decision. Each one is yours alone.

Configured for Your Office Before It Arrives

Every Obsidian Sovereign is hand-configured before delivery. Every component is selected for its category, not its price point. If your office runs particular software, a particular security stack, or particular workflows, those requirements are built in before the unit leaves our hands. The system bends to your office. Your office does not bend to the system. Available Q3 2026. Private briefings accepted now.