Built in Switzerland · Private by architecture · Neutral by jurisdiction

A sovereign workspace for your most sensitive work — confidentially and collaboratively.

A private, shared workspace for your whole team — where nothing leaves, and the group's knowledge compounds year after year instead of scattering across inboxes and a few people's heads.

For a century, the world trusted Switzerland to hold what mattered most — discreetly, neutrally, beyond easy foreign reach. That trust was built on banking. The asset has changed: today it is data, and the AI run over it. The instinct, the neutrality, and the law remain — now expressed as some of the world's strictest data-protection rules, sovereign infrastructure on Swiss soil, and AI governed end to end under Swiss law.

The vault was built for wealth.
The one that matters now holds data.
Try the live demo →
Built in Zug · Hosted on Swiss-sovereign infrastructure

Built for groups whose work shares the same shape: closed, sensitive, and worth getting right.

Every closed group of professionals — five directors on a board, a dozen people in a deal room, a legal team running a regulated matter — is now caught in the same bind. The consumer AI products are out of bounds: feeding privileged or material non-public information to a vendor whose terms permit training is not a defensible decision. The enterprise AI products are built for thousands of seats inside a single company, not for the small, cross-organisational, advisor-laden groups where the actual decisions get made — and they tie the work to AI run by companies, and under jurisdictions, that the group does not answer to.

So most of these groups do nothing. Members go back to their personal AI accounts on the side, and the work of the group continues to live where it has always lived — in inboxes, scattered files, and the heads of two or three people.

Meanwhile, what the group knows lives everywhere
and what it decided lives in no one place at all.

None of the AI on offer was built for work like this — closed, sensitive, shared across organisations, and meant to last. Consumer tools cannot be trusted with it; enterprise tools do not fit it, and tie you to AI you do not control; and in neither does the group's knowledge ever compound.

So we built the one that does.

Hosted in
Basel data centre
Governed by
Swiss Federal Data Protection Act (FADP)
Data residency
Never leaves Switzerland by default
Operated by
A Swiss team, accountable to Swiss law
Need no-knowledge custody
Our Dedicated tier adds customer-controlled HSM keys and confidential computing in production — neither our engineers nor the infrastructure operator hold a copy of your decryption keys. See pricing.

A short list of teams we're built for.

Steward is shaped for groups where every member is trusted, every document is sensitive, and the work persists across years. If your team looks like one of the four below, the fit is likely. If it looks like something we have not yet listed, we want to hear about it — that prompt is at the bottom of this page.

01

Boards & secretariats

Listed companies, foundations, regulated entities, pension boards.

Shape
5–15 directors plus a small secretariat. Members rarely overlap with day-to-day staff. Materials cycle quarterly.
The work
Board books, minutes, resolutions, regulator filings, committee papers — and the long memory of how the board has actually decided things over time.
What hurts
Diligent and Boardvantage hold the documents but cannot reason over them. ChatGPT cannot be used for material non-public information.

What did the board actually decide on the dividend policy review of 2022, and what conditions did the audit committee attach?

Try the board-style demo
02

Transaction & investment committees

M&A teams, deal rooms, IC, private equity, special situations.

Shape
A core team of 4–10 with a wider ring of advisors — counsel, accountants, bankers — who come and go by deal.
The work
Diligence files, IC memos, term sheets, integration plans, post-mortems. Cross-deal pattern recognition matters as much as the live deal.
What hurts
Each deal starts from a blank slate. Lessons from the last three deals live in two people's heads. NDAs prevent any consumer AI use.

Across our last twelve diligences, where did revenue quality assumptions diverge from what we found in the first ninety days post-close?

Try the deal-team demo
03

Legal & matter teams

In-house legal, GC offices, outside counsel project rooms, regulatory teams.

Shape
A matter team formed for a specific case or workstream — internal and external counsel, often across firms and jurisdictions.
The work
Pleadings, contracts, opinions, regulator correspondence, e-discovery review. Privilege and confidentiality are not features — they are the floor.
What hurts
Discovery platforms answer search queries but not actual legal questions. Most AI tools' terms of service are incompatible with privilege.

What positions has opposing counsel taken on indemnification scope across the last four matters with this firm — and where have they conceded?

Try the legal-matter demo
04

Family offices

Single- and multi-family offices, principals, and the advisors around them.

Shape
A small core team around a family or principal, plus a rotating ring of advisors — counsel, bankers, tax, trustees — across jurisdictions and generations.
The work
Investment files, entity and trust structures, board and committee papers, correspondence — and decades of history that nobody has had time to make searchable.
What hurts
The family's memory lives in a few trusted heads and scattered files. Every new advisor or next-generation member starts from zero. Consumer AI is unthinkable for assets this private.

When we restructured the family's holding companies in 2019, what did each advisor recommend — and what did we decide, and why?

Try the family-office demo

Five capabilities. One workspace, built to compound.

Each works on its own. Together they make Steward the one place your whole team works — where the group's knowledge compounds across years, not a tool you log into when you remember to.

01

Institutional memory

Ask anything that lives in your group's documents, decisions and prior conversations. Steward reads across your permitted corpus and answers with citations to the actual source — minutes, memos, term sheets, opinions, board packs, advisor correspondence. Your group's institutional knowledge becomes searchable, not just storable, and every answer can be promoted into the group's standing record.

02

General-purpose AI, in confidence

The everyday questions your members would otherwise put to ChatGPT or Claude — research, drafting, summarising, structuring — answered inside your sovereign workspace. Frontier model quality with the data protections your work requires. No queries leave the group by default. No conversation enters anyone's training set.

03

Agents that run on schedule

Steward can take on recurring work and run it for you. Compile the pre-read for every quarterly meeting from the prior period's documents. Pull every newly-filed document for an active matter and summarise the changes. Flag anything overdue from the last meeting's resolutions. Set the task once. The work happens, with audit trail, until you change your mind.

04

A knowledge graph that connects everything

Underneath every answer, a living map of your group: people, entities, matters, decisions, advisors, deadlines — linked by how they touch each other. The graph is what lets Steward navigate your records the way a long-serving secretary would, rather than searching them like a database. Visible in the workspace, exportable as your own folder of wiki pages — yours regardless of where your stack goes.

05
The platform layer

A foundation you can build on

Steward is the foundation, not the ceiling. Reporting dashboards, custom workflows for your particular committee, integrations with the systems of record you already run — anything you'll want to add over the next five years reads from the same sovereign substrate. Build it yourselves on the same files Steward has been gathering all along, or commission us to build it for you. Either way, no re-ingestion, no re-migration, no vendor lock-in at the foundation layer. The technology decision you make once, that makes every subsequent decision easier.

Most providers ask you to trust them. We have built Steward so you don't have to.

Consumer AI
ChatGPT, Claude, Gemini
Enterprise AI
Copilot, ChatGPT Enterprise
Steward
 
Frontier-model quality
No training on your data (under contract)
Available individually, instantly
Centralised administration (SSO, audit logs)
Hosted in Switzerland, under Swiss law
Vendor's engineers cannot read your data (Dedicated tier)
Outside US CLOUD Act reach
Built for closed groups across organisations
Knowledge graph across the group's documents
Where does my data live?
In Switzerland. In a Basel data centre, operated by a Swiss team, governed by Swiss law. It does not leave the country.
Can your staff read it?
On our Hosted tier, a small audited team of ANA Wealth operators has database access for support and incident response — every access event is logged. On our Dedicated tier, customer-controlled keys and confidential computing mean no Steward or Phoeniqs employee has a technical path to your data, even in principle. The full Hosted-tier disclosure is at /data-handling; the engagement shape is at /pricing.
What if a court orders you to hand my data over?
A Swiss court order against ANA Wealth or Phoeniqs would be required to compel disclosure. The U.S. CLOUD Act does not reach Swiss-owned infrastructure, and non-Swiss authorities must go through mutual legal assistance treaties via Swiss courts — a higher bar than a subpoena in their own jurisdiction. On our Dedicated tier, the keys to decrypt your data are held by you; even a Swiss court order to us would not produce readable data.
What happens when we want to leave?
You export everything — every document, every note, every audit log entry — as ordinary files in standard formats. No re-licensing. No vendor handover process. Walk away with what is yours.

Confidential computing — what the Dedicated tier rests on.

When the Dedicated tier guarantees that no one outside your group can read your data — not us, not the infrastructure provider, not even under a court order to either of us — that claim has to rest on something more than policy. It rests on a category of infrastructure called confidential computing: memory that stays encrypted even while it is being processed, and hardware that attests to exactly what code is running before it agrees to unlock your data. On Hosted, encryption at rest under platform-managed keys is the protection; on Dedicated, the addition of confidential computing makes "we cannot read your data" a property of the silicon, not a promise.

01

What it requires

  • CPUs and GPUs with hardware memory-encryption (AMD SEV-SNP, Intel TDX, NVIDIA Confidential H100)
  • Cryptographic attestation that proves which code is running inside the encrypted boundary before any data is released to it
  • A key custody arrangement where neither the infrastructure operator nor the application vendor holds a copy of your decryption keys
02

Where it is offered in production

  • A small number of major cloud providers globally — perhaps four or five, no more
  • A much smaller number that combine it with Swiss data residency
  • A smaller number still that combine it with full Swiss legal jurisdiction — no US CLOUD Act exposure, no foreign-court reach into the data while it is being processed
03

What we use

  • An established Swiss cloud service provider that serves Swiss banks and government agencies under Swiss law
  • Production confidential computing across the inference and storage path — not a beta feature, not a roadmap item
  • Tenant-held keys, generated and rotated by your administrator. We never receive a copy. Neither does the infrastructure provider

This is the substrate underneath the Dedicated tier's no-knowledge guarantee. Without it, "no one but you can see your data" is something you have to trust. With it, it is something you can verify.

A shared sovereign service. Dedicated when posture requires it.

Two tiers: a multi-tenant Swiss-sovereign Hosted service for most engagements, and a single-tenant Dedicated option for organisations whose posture requires customer-controlled keys and confidential computing. Pricing is per-seat with a tenant base; numbers come from a short scoping conversation.

See the engagement shape

A note about this page

We are still learning
which teams recognise themselves here.

Steward began inside a Swiss family office, where the fit was unambiguous. As we extend to other closed groups, we are deliberately listening before we generalise. The four personas above are our hypotheses about which teams share enough of the underlying shape to make Steward worth their consideration. We may be missing one — or describing one in language that does not match how the work feels from the inside.

If you are inside a group that the four cards above do not quite capture, we would value a few sentences from you about the shape of your work.

Book a call about your team

We read each one personally. Replies within a few days.

Built first for the group we know best.

Steward was built inside ANA Wealth, a Swiss family office, for our own work. Family offices were the original group: small, trust-bound, document-heavy, advisor-laden, with decades of history that nobody had time to make searchable. The product still serves them — and a separate page describes that fit in detail.

Extending Steward beyond family offices is a deliberate move, not an afterthought. The substrate underneath — sovereign Swiss infrastructure, knowledge graph plus vector retrieval, House Rules per workspace — turned out to fit a wider class of work than the original use case. Boards, deal teams, legal matters, and strategy committees share the same shape: closed membership, sensitive material, multi-year continuity, and a need for AI that cannot be answered by either consumer or enterprise products as they stand.

We are extending carefully, with one director or general counsel at a time, because the groups we serve are not the kind of groups that tolerate being a beta market.