We are building the agentic desktop workplace for everyone whose control over agents and data is mission-critical.

Made in Europe.
First release planned for 2027.
See what is coming, and be part of it from the start.

For teams and topics that are too confidential for today's AI tools.

Across Europe, office workers with strictly confidential work get limited or no access to AI.
Sondy is being built to give them the agentic experience they otherwise know only from their private machines.

Glass office tower, seen from below

Private sector

Examples:

  • Critical infrastructure
  • Regulated research
  • Hospitals and clinical research
  • Defence and space
  • Law and audit firms
  • M&A teams
  • Patent attorneys
Public building with a German flag

Public sector

Examples:

  • Security authorities
  • Parliaments
  • Ministries
  • Courts
  • Armed forces
  • Intelligence services
  • Police
  • Foreign service
  • Crisis management

Truly agentic, across your applications.

Say what you need once, and the agent carries it through your applications and the operating system itself.
A complete workspace, designed for the future of agentic work.

Example use cases:
“Read last week's mail on the Miller deal, find the latest offer in our files, put its new figures into the client memo, check the source links, and draft the summary mail to the client for me to review.”

Central AI interface

Realtime security monitoring

Examples:

  • No client mail sent to a cloud model
  • No client files leaving the workstation
  • No document sent to a cloud outside the EU
  • No compromised agent sending data out
  • Mail
  • Files
  • Documents
  • Web browser
“Analyse last week's data from the lab system, post the three biggest outliers in the team chat, collect everyone's explanations on the idea board, and schedule a review for the strongest one.”

Central AI interface

Realtime security monitoring

Examples:

  • No unpublished results sent to a cloud model
  • No results shared outside the lab team
  • No ideas sent to a cloud outside the EU
  • No meeting details visible outside the team
  • Data analysis
  • Team chat
  • Idea board
  • Calendar
“Translate the seized chats, check the names and numbers against the police databases, draw the link chart of who was in contact with whom and when, and draft the report for the case file.”

Central AI interface

Realtime security monitoring

Examples:

  • No case data sent to a cloud model
  • No personal data sent to a cloud outside the EU
  • No name looked up outside the police databases
  • No evidence leaving the workstation
  • Translation
  • Police databases
  • Link chart
  • Case file

Control that goes as deep as the agents do.

We combine security safeguards at OS and machine level. Each is designed for agents that move down the stack, from single apps into the operating system itself. Together they close gaps that other approaches leave open.

1Controlled actions

Every step an agent takes is checked first: it is allowed, asked about or refused.

  • Allow: the action runs within your policy.
  • Ask: the user confirms first.
  • Refuse: the action never runs.
  • Audit log: a full record of what the agent did, kept to strict user privacy standards.

2Clean, sealed, reset

Technically an immutable, stateless system. On request fully local-first and air-gapped.

  • Clean: one approved, fingerprinted and audited setup, checked at every start.
  • Sealed: one connection to the outside, every request on it passes your policy.
  • Reset: back to that setup, like a factory reset, after each project, at a reboot, or at the CISO team's button.

3Future-proof

Agents will soon run entire workstations, deeper than any app reaches.

  • Today: agents work in the cloud and inside single apps.
  • Next: they run the whole workstation, down to the operating system.
  • In the OS: our control sits in the operating system itself, as deep as agents will go.

No fleet migration required.

We build for the people who work with Sondy every day, and just as much for those who roll it out, maintain it and manage procurement. It is being built to start in a window, on the machines your IT already manages.

Other operating system

It runs like an app

An agentic workspace, being built for the highest security, in a window on the laptop you already have.

  • Apps
  • System
  • Kernel
  • Boot

Or as the only operating system

For those who need the extra level of security: Sondy is being built to secure the whole workstation, from boot and kernel up.

Built for trust

  • Regular independent audits: outside security firms test what we ship.
  • An open, reviewable core: anyone can review our code in our public repository.
  • Open standards instead of lock-in: we contribute our protocol between agents and applications to the Linux community.
  • Local and privacy first by design: runs on your own hardware, no vendor connection needed, bring your own models.
  • Our certification targets: CE mark under the EU Cyber Resilience Act, EUCC, BSI approval for VS-NfD.CE markingEUCCBundesamt für Sicherheit in der Informationstechnik (BSI)
  • A sovereign stack, made in Europe: GDPR compliant, on European infrastructure.Flag of the European UnionGDPR

Coming soon.

There's more to talk about. Let's get in touch.

InvestorsInvest while we build it.Product development has started, and early investors are welcome.Meet the team and experience a live demo.Get investor accessPeopleWork on hard problems that matter.How do you audit everything an agent does without watching the person it works for?How do you evaluate an agent on data that nobody is allowed to see?How does a person stay in control of work they no longer do by hand?How do you control agents without limiting their intelligence?Who builds that in Europe, on open source?See our open positions
Expert communitySee it first. Then break it.Private preview planned for early 2027.For experts in engineering, AI, Linux, security and IT.Join the waitlist