on-premise
An AI assistant that keeps data inside the company
The company map and the assistant for customers and team — on your servers and your models. No cloud, no outbound connections by default, documentation for your security team.
- your servers
- local models
- FZ-152
- security documentation
who it's for
For companies that can't use the cloud
Banks and fintech
Customer data and conversations never leave the perimeter. The assistant works in internal systems with employee-level permissions.
Industry
Orders, supply, QC and paperwork — inside the plant's closed network, including with no internet access.
Public sector and large groups
One installation for the group, sign-in via your corporate directory, a log of every assistant action.
what's inside
The same platform as in the cloud
The box is built from the same code. The only differences are where the cloud reaches outside — those are switched off.
Company map
Departments, processes and systems, leaks in rubles and the autonomy index.
Assistant for your team
Works in your systems, acts only after your OK, with before → after.
Assistant for customers
Site widget and internal channels — as your administrator decides.
Company knowledge
Policies and documents with sections and access rights; search only within what's allowed.
Speech recognition
Voice messages and calls are transcribed by a local model inside the perimeter.
Administration
A dedicated admin panel: users, models, logs, updates.
security
What your security team will see
We hand over the security description before installation: components, network interactions, data, license, updates.
No outbound connections
By default the box contacts nobody — neither us nor third parties. No telemetry, no online license checks.
Egress guard
Any connection outside the perimeter is blocked until an administrator explicitly allows it.
Sign-in via your directory
LDAP / Active Directory and Keycloak. No open registration.
Signed bundle
Images ship as an archive with checksums and an Ed25519 signature — verify before installing.
Least privilege
Containers run as non-root with no Linux capabilities; the only entry point is 443.
No phone-home license
The license is a signed file. When it expires, the system goes read-only and no data is lost.
models
Models stay inside the perimeter
The assistant runs on the model you host. Cloud models are never called on-premise.
Company model server
Ollama, vLLM or LM Studio on your servers — connected by address in settings.
Models on employees' computers
An account can use its own local model — only from addresses your administrator allows.
Model trials
Before launch we test how the chosen model handles your tasks.
how it goes
From request to production
Call and security requirements
Goals, systems, infrastructure and documentation requirements.
Trial installation
The bundle on your test bench, security review, model trials.
Audit and map
The company map on your data: where manual work is and what to hand to AI first.
Launch and support
Connecting systems, training, updates with database backup and rollback.
pricing
Annual license
The price depends on active users and the scope of rollout. You get an exact offer after a call.
- Base license
- from 1,800,000 ₽ per year
- Active user
- 6,000 ₽ per year
- Rollout
- from 1,500,000 ₽
request
Discuss on-premise
Tell us about your company — we'll prepare an offer, the security description and a trial bundle.