Strong AI teams. Private by design.
Secure, Managed Open-Weight AI Teams and Workflows in a Private Cloud. – Bring Your Private Data Securely Into the AI Era (how we do it?).
Give your AI team a job — research, process data, create content, update systems, and run workflows across your business. Powered by leading open-weight models from Qwen, Mistral, and GLM* in a secure private cloud, where your data stays fully private and under your control. Manage work, get notified, and talk with your AI team from the VAKEVA mobile app.
* GLM models are coming soon.
One workflow, end to end
- Data in, on your terms
- An API or your own store, read-only or writable, strict or loose — the agent gets exactly the right you granted.
- Roles, not prompts
- Each step is done by a role with its own instructions and skills, on the model best suited to it. You write the brief; the team does the work.
- Two endings, every run
- The result is saved in 1 GB of storage only you can decrypt, and acted on — posted, sent, written back. Every access is in your audit log.
Teams of agents
Compose a team from proven roles — engineer, analyst, writer, reviewer and more — each paired with the open-weight model best suited to the job. The system suggests the fit.
Scheduled workflows
Ordered steps, each with a clear brief, linked into one flow that runs daily, weekly, monthly or once a year — and reports its success rate and time to completion.
Your own data, on your terms
Attach external APIs or internal stores, strict or loose, input or output. Nothing is used in a way you did not authorise, and you can see that it was not.
Read how
Bring your private data securely into the AI era
Most AI tools ask you to send your customer lists, contracts and numbers to a public API and trust a policy page. We built VAKEVA the other way round: the models come to your data, on infrastructure we run ourselves, inside rights you grant one by one, with a record of every access you can read. Secure and auditable is not a feature we added — it is the shape of the product.
- Our own servers, in the European Union
- Your agents run on hardware we operate in EU data centres, not on a public model API. No prompt, no document and no result leaves the EU, and no third-party model provider ever sees your data. That is what makes the data-residency question a one-line answer for an EU company.
- Open-weight models, never trained on you
- Qwen, Mistral and soon GLM* are open-weight models we host and freeze at a known version. They cannot phone home, they are not improved on your conversations, and your data is never used to train anything — ours or anyone else's.
- Encrypted storage only you can decrypt
- Every team gets 1 GB of storage encrypted at rest with a key held for that team alone. Our own operators cannot read what your agents saved. Delete the team and the key goes with it — which is how the right to erasure becomes an action, not a ticket.
- Rights, granted one at a time
- A data source is attached to an agent as a right: read or write, strict or loose, input or output. An agent without the right cannot reach the data — the runtime refuses, it does not ask the model to behave. Least privilege is the default, not a checkbox.
- An audit log in plain language
- Every access is recorded with the agent, the workflow, the step and the right that made it, and shown to you as a sentence a compliance officer can read. Article 30 records of processing stop being a spreadsheet you maintain by hand; they are a page you open.
- Isolated per team, from the runtime up
- Each team's agents run in their own isolated runtime with their own keys and quota. Nothing is shared between customers but the machines, and the machines never keep your data once a run is finished.
GDPR-ready for EU companies
- You are the controller, we are the processor. A data processing agreement is part of every plan, with the sub-processor list kept short and published — hosting only, no model vendors, no analytics resellers.
- Data stays in the EU. Processing, storage and backups are in EU data centres, so no transfer mechanism is needed for the data your agents touch.
- Purpose limitation, enforced. An agent can only do what its role and its granted rights allow, so the data is used for the purpose you set up and the log proves it was used for nothing else.
- Rights of the data subject, in one place. Access, rectification and erasure requests are served from the encrypted store and the audit log — the same records you already see, so an answer takes minutes, not a project.
- Security by design and by default. Encryption at rest and in transit, per-team keys, least-privilege rights, two-factor login for administrators, and no public model API in the path.
- Ready for your DPIA. Because every data flow is declared as a right and recorded as an access, the impact assessment your DPO asks for describes a system that already documents itself.
VAKEVA is pre-launch; the DPA and sub-processor list are published with the first paid plans. Pay for runtime minutes per model, billed monthly, from 120 minutes a week. * GLM models are coming soon.