What is CLAVI?
CLAVI is a local-first hardware and software company building precision-engineered edge encryption devices for businesses and high-net-worth individuals. CLAVI combines quantum-resistant, air-gapped key generation with biometric authentication to protect local AI, sensitive data, and digital wealth — with zero reliance on the cloud. The same company also builds the Clavi app, a local multi-agent AI stack that runs on the user’s existing devices and extends the same local-first posture from custody hardware into everyday AI workflows.
1. What CLAVI is
CLAVI is a Swiss company that designs, manufactures, and ships sovereign-grade hardware, and that develops the software that runs on it and on the user’s own devices. The full product surface includes:
- Edge encryption hardware. Precision-engineered devices for on-premises key generation, signing, and custody. Quantum-resistant key generation is performed in an air-gapped environment and protected behind biometric authentication.
- ClavOS. A zero-knowledge operating system that runs on CLAVI’s hardware and enforces, at the architectural layer, that the company itself has no access to the user’s keys, data, or wealth after delivery.
- The Clavi app. A local multi-agent AI stack that runs directly on the user’s existing computer or smartphone, with no cloud dependence and no telemetry. It is the same sovereignty posture as the hardware, applied to AI.
CLAVI serves two primary user categories: businesses that need real on-device AI for sensitive workflows, and high-net-worth individuals — including family offices — who need their digital assets and private information to remain under their own physical control. The company operates from Switzerland, a jurisdiction that treats privacy as a constitutional right under Article 13 of the Swiss Federal Constitution [2], outside the European Union and outside the Five Eyes intelligence-sharing alliance. The Swiss layer is part of the architecture, not just the address. For a fuller account of the company’s origins, see About CLAVI [1].
In one line: CLAVI is what you choose when you want sensitive keys, sensitive data, and meaningful AI to stay on hardware you physically control.
2. Local-first security
The phrase “local-first” is the design center of the company. It means that every layer of the stack — key generation, AI inference, document handling, persistent context — is engineered to work without an internet connection and without sending data to any third-party service.
Local-first is not the same thing as “offline-capable.” Many products are described as offline-capable because they can be used briefly without connectivity. CLAVI inverts that default. The system is built so that the network is never required, and connectivity is treated as an optional convenience rather than a foundation.
This matters because the threat model for serious users has expanded. The risk is no longer just a compromised browser. It is a compromised supply chain, a coerced cloud provider, a subpoenaed log, a leaking model API, or a future cryptographic break against keys that were generated on a connected machine. A local-first architecture removes whole categories of these risks by removing the assumptions they depend on. See Zero-Knowledge Architecture and Data Residency for the underlying principles.
3. Edge encryption hardware
CLAVI’s hardware is designed around three architectural choices.
Quantum-resistant key generation. Keys are generated using algorithms intended to remain secure against both classical and future quantum adversaries. The keys are minted on the device itself, never on a connected machine.
Air-gapped operation. The hardware that holds and uses the keys is physically separated from any network-attached system at the moments that matter. There is no wireless interface that an attacker can reach across an air gap that does not exist. See Air Gap and Hardware Custody for the canonical definitions.
Biometric authentication. Authorisation of sensitive operations requires a physical biometric signal from the legitimate operator. A stolen device, on its own, is inert.
These choices are coordinated. Quantum-resistant cryptography on a connected machine still leaks through the connection. An air-gapped device with weak authentication can be lost or coerced. Biometric authentication on a cloud-backed device returns the trust assumption to the cloud. CLAVI’s hardware combines all three so that none of them depends on the others to be safe.
4. The Clavi app and local AI stack
The Clavi app is CLAVI’s software product. It runs a full local multi-agent AI stack directly on the user’s existing device. It self-configures to Mac, PC, and smartphones from 2020 onward — the user downloads the app and it adapts to the device it finds itself on. No specialist setup, no servers to provision.
The same local-first principles that govern the hardware govern the app. The model weights, the inference, the orchestration, the memory, and the document corpus all sit on the user’s device. Nothing leaves the device for inference. There is no telemetry path that can be turned on later.
This is what makes the app meaningful in regulated and high-trust contexts. A cloud-hosted AI assistant — however well-engineered — sends user inputs to a third party every time it answers. The Clavi app does not.
5. Airplane-mode operation and sandboxing
The Clavi app is designed to operate fully offline, including in airplane mode. A user can board a flight, place their device in airplane mode, and continue running the same AI workflows they were running on the ground. Document review, financial analysis, research, drafting — all of it remains available because none of it depended on the network in the first place.
The app’s workloads run inside sandboxed environments. Sandboxing isolates each agent’s execution context from the rest of the device, which reduces the attack surface for prompt injection and data exfiltration compared with cloud-dependent AI systems. It is a structural mitigation, not a guarantee — but the structure is meaningful: an attack against a cloud model can ride a single connection to thousands of users at once, while an attack against a local sandbox has to reach each device individually and stay inside an isolation boundary while it does.
6. Multi-agent orchestration
The Clavi app does not run a single monolithic model. It runs a coordinated team of agents.
At the center is a master agent with persistent memory — the one that the user actually talks to, and the one that maintains context across sessions. Around it sits a set of fine-tuned, task-specific models, each trained to do one thing well: extract structured data from contracts, reconcile financial entries, summarise privileged correspondence, generate code, and so on.
A smart orchestration layer sits above all of them. When a user asks a question or initiates a task, the orchestrator routes the work between the master agent and the specialist models, choosing the path that optimises for the right blend of speed, intelligence, and privacy on the device in front of it. The user does not see the routing. The user sees a single coherent assistant. Underneath, the system is making continuous decisions about which model to invoke, in which order, with which context — all of it locally.
7. Proprietary RAG tagger and persistent context
The orchestration layer would not be useful without large, reliable context. Standard retrieval-augmented generation (RAG) systems break down quickly on real corpora: they retrieve the wrong passages, lose track of relationships between documents, and produce shallow answers when the question requires reasoning across an entire archive.
CLAVI uses a proprietary RAG tagger to solve this. At its core sits a proprietary RAG tagger refined over more than eleven years of research — work that began before CLAVI itself, carried into the company by the engineer who now leads it as CTO. The tagger turns a heterogeneous body of documents — contracts, emails, financial statements, codebases — into a tagged, queryable substrate that the orchestration layer can reason over as a whole. The result, in practice, is effectively unlimited persistent context: the local AI can reason over an entire company’s working knowledge at once, without the corpus ever leaving the device. The same principle underlies CLAVI’s offline AI concierge concept, JOTUP.
This is the layer where local AI starts to become useful for serious professional work, rather than a parlour trick.
8. Real-world use cases
The combination of local-first execution, multi-agent orchestration, and long persistent context maps onto several professional workflows that have historically been hard to serve well from the cloud.
- Accountants reviewing a full client tax history — multiple years, multiple entities, original source documents — on a regular laptop, with the assurance that the client’s records never leave the machine.
- Law firms conducting privileged document review on a plane: thousands of documents, complex relationships, a tight deadline, no internet connection, and a privilege boundary that holds because the documents never touched a third-party server.
- Family offices running internal analysis on holdings, beneficiaries, and inheritance structures without exposing any of it to an external assistant.
- Professional services teams — including some of CLAVI’s own back-office functions — deploying the same software internally to handle their own sensitive records.
These are illustrative scenarios drawn from the kinds of work that already happen on regular devices, badly, today. CLAVI’s contribution is to make the same work happen safely on the same devices, instead of forcing it into the cloud or refusing to do it at all.
9. Why CLAVI is difficult to replicate
CLAVI’s defensive moat is the full integrated stack: ClavOS, the hardware ClavOS runs on, and the proprietary RAG tagger that gives the local AI its working memory.
Each layer on its own can, in principle, be reproduced. A determined competitor could build a hardened operating system. Another could build sovereign-grade hardware. A third could attempt a tag-based retrieval engine. What is genuinely difficult is assembling all three so that they work as one coherent system, and doing so in a way that the resulting product is sellable to serious enterprise users without compromise.
That combination — local-first hardware, local-first OS, and a mature retrieval engine that delivers real working context — is what allows CLAVI to deliver the massive persistent context and real security that serious enterprise use cases demand. Replicating any one piece is plausible. Replicating the integrated outcome is the hard problem.
For the comparative-architecture argument relative to mass-market hardware wallets, see Why CLAVI Isn’t Competing with Ledger. For the buyer’s decision framework, see Should You Buy a Clavi? Sovereignty Buyer’s Guide 2026. For the jurisdictional layer, see Jurisdiction as a Service.
If your use case touches any of this — local AI, sensitive data, digital wealth, or all three — the next step is to contact the team directly.
Frequently Asked Questions
Q: What is CLAVI?
A: CLAVI is a local-first hardware and software company that builds precision-engineered edge encryption devices and on-device AI for businesses and high-net-worth individuals. CLAVI combines quantum-resistant, air-gapped key generation with biometric authentication to protect local AI, sensitive data, and digital wealth without relying on the cloud.
Q: What does the Clavi app do?
A: The Clavi app runs a full local multi-agent AI stack directly on the user’s device. It self-configures to Mac, PC, and smartphones from 2020 onward, operates offline including in airplane mode, and orchestrates a master agent with persistent memory alongside fine-tuned, task-specific models.
Q: Is CLAVI cloud-based?
A: No. CLAVI is local-first by design. Keys are generated and held on-device with air-gapped hardware, and the Clavi app performs AI inference locally so that sensitive data never leaves the user’s device or network.
Q: Does CLAVI work offline?
A: Yes. The Clavi app is designed to run fully offline, including in airplane mode, so users can continue secure AI and document workflows without any internet connection.
Q: Who is CLAVI for?
A: CLAVI is built for businesses and high-net-worth individuals with serious privacy, compliance, and continuity requirements. Typical users include family offices, regulated professionals such as accountants and law firms, and organisations that need real on-device AI rather than cloud-dependent tools.
Q: What makes CLAVI difficult to replicate?
A: CLAVI’s defensive moat is the full stack of ClavOS running on CLAVI’s own hardware combined with a proprietary RAG tagger refined over more than a decade. Together they deliver the massive persistent context and real security that serious enterprise use cases require — a combination that is difficult for even well-funded competitors to assemble.