An AI agent that keeps your data inside
Doka deploys inside your perimeter: local inference, no cloud, no accounts. For companies where cloud AI is banned by security policy.
Problem
Cloud AI hits the security policy wall
Data can't go to the cloud
Banking secrecy, personal data, state secrets, attorney-client privilege — regulation forbids sending them to an external service.
Shadow usage
Employees paste documents into public chatbots anyway. No control, a leak is a matter of time.
Import substitution
You need solutions on local models and infrastructure, not foreign SaaS with unclear jurisdiction.
Solution
Doka on-prem — an agent inside the perimeter
No data leakage
Inference runs locally, on the device or your server. Requests and files never leave the perimeter.
White-label
Your brand, your name, your default policies. Doka embeds as an internal company tool.
Local models
Integration with GigaChat and YandexGPT, plus any OpenAI-compatible and local models.
Access control
Access to files, terminal and network only by explicit permission. Security sets the policy, not the user.
Rollout
How we deploy
Pilot
We install Doka for a limited group, connect your models and data, validate scenarios.
Deployment
Mass install via GPO or MDM, policy and integration setup for your infrastructure.
Support
SLA, updates, staff training and support under the chosen package.
SLA packages
Support packages
Basic
Small teams and first rollout.
- ◆On-prem deployment
- ◆Email support, 1 business day
- ◆Updates and patches
- ◆Up to 50 seats
Business
Department or branch rollout.
- ◆Everything in Basic
- ◆GPO/MDM deployment
- ◆GigaChat / YandexGPT integration
- ◆Priority support, 4h response
- ◆Up to 250 seats
Enterprise
Scale and white-label.
- ◆Everything in Business
- ◆White-label and custom policies
- ◆Dedicated manager, contractual SLA
- ◆Staff training
- ◆Unlimited seats
◬ Free pilot
We run a pilot at no cost in exchange for an anonymized case. Then you move to one of the packages.
SLA packages
Request a pilot
Tell us about the task — we'll reply and propose a pilot format.