RESTART AI Enterprise Platform is the enterprise AI platform of RESTART Group: a single AI environment within the company perimeter: language models (LLM), AI agents and a knowledge base work on top of a common core, and employee access rights and the request and response log remain under the control of the customer. The platform is deployed on-premise, in a private cloud or hybrid; 10 corporate modules are connected to the core - from document search to service desk and contracts.
RESTART AI Enterprise Platform is the corporate AI platform of RESTART Group for large companies: language models, AI agents and a knowledge base in the customer’s context, with access rights, log and human verification.
An enterprise AI platform is not a separate chatbot, but a common layer on which a company runs all AI scenarios. Employees work with AI in one controlled loop, and each service knows its data sources, user roles, access restrictions and process owner. The new scenario connects to an already created core, rather than being built from scratch.
Model layer: LLM connection, prompt versions, limits and cost of requests. The model is selected according to the task and data requirements.
AI agent designer, action scenarios, business rules and human verification for critical operations.
RAG: document loaders, vector indexes, answers to authorized sources with a link to the document.
Users, roles and rights: the employee does not receive in response what he does not have access to.
Actions, requests, responses, sources, prompt versions and control events are recorded - for auditing and information security.
Connectors to 1C, SAP, ERP, CRM, EDMS, BI, DWH, Service Desk, GitLab, Jira, Confluence and internal API.
The architecture is selected according to the customer's requirements for data and information security. The placement, access model, environment composition, backup, SLA and procedure for uploading or deleting data are fixed in the technical specifications and contract.
The platform and language models run on the customer’s servers, inside its perimeter. Data does not go to external services.
Isolated loop with data transfer restrictions, logs, secrets management and customer information security regulations.
Part of the environment is at the customer’s, part is in a private cloud. External LLM providers are connected only according to an agreed architecture.
RESTART can deliver the platform along with AI-driven infrastructure: development, testing and production loops, GPU, storage, backup, monitoring and maintenance.
Typical platform outline: web interface, backend API, PostgreSQL database, vector storage, file storage, task queue and handlers, LLM adapters, GPU node and monitoring. Below are resource guidelines for each environment; the specific configuration depends on the number of users, frequency of requests and selected models and is recorded in the technical specifications.
| environment | Purpose | Example resources |
|---|---|---|
| Development | Development, debugging, experiments | 4–8 vCPU, 16–32 GB RAM, 150–300 GB SSD, GPU as needed |
| Test / Staging | Pilot, checking releases and integrations, acceptance | 8+ vCPU, 32+ GB RAM, 300+ GB SSD/NVMe, L4 / A10 / L40S or task equivalent |
| Production | User experience, API, AI chat, RAG, knowledge bases | 16+ vCPU, 64–128+ GB RAM, 1+ TB NVMe, L40S 48 GB / A100 80 GB or equivalent as agreed |
| Backup / Storage | Backups, source documents, logs | Separate storage environment with regulations for storing and unloading data |
The GPU is not always needed: it is required for local LLM, embedding, reranking, processing large amounts of documents and high-load scenarios. For some scenarios, a CPU environment or external LLM providers according to an agreed architecture are sufficient.
Example from bank case: pilot worked on 2 × A5000 24 GB (2 parallel threads to the GPU), production in the internal loop - on 4 × A5000 24 GB (6 threads); 2 × A100 80 GB are provided for external users, parameters are specified after load testing.
When hosted on-premise or in a private cloud, documents, requests and responses remain within the customer’s perimeter. Transfer of data to external providers is limited by architecture and contract.
Answers are based only on permitted data, taking into account the employee’s role; sources are connected according to an agreed list.
Each request, response, source used and prompt version are recorded in a log - for analyzing incidents and checking information security.
For critical processes, the agent’s action is confirmed by an employee; Agents' powers are limited - read more about secure AI automation.
Work with PD according to Federal Law No. 152-FZ begins at the survey stage: sources, access rights, journals, placement; if necessary - data masking.
Access control, secrets management, backup, monitoring and additional information security measures according to customer requirements.
RESTART AI Enterprise Platform is needed where AI can no longer be launched as a set of disparate experiments. In a large company, each AI scenario must know its data sources, user roles, access restrictions, logging rules, information security requirements and process owner.
The platform collects this into a manageable loop: AI agents, RAGs, models, prompts, integrations, auditing, request costs and application modules do not work separately from each other, but on top of a common core.
AI scenarios evolve through unified roles, sources, logs, policies, and quality control.
Each new module is connected to an already created platform, rather than being built from scratch.
Access rights, auditing, source restrictions, placement and work with personal data are taken into account at the architectural level.
The platform communicates with ERP, 1C, SAP, BI, DWH, EDMS, Service Desk, GitLab, Jira, portals and knowledge bases.
Users, roles, access rights, logs, auditing, model layer, prompts, limits and request costs.
RAG, document loaders, vector indexes, source citation, corporate knowledge base search.
AI agent builder, action scenarios, human verification, business rules and secure automation.
Connectors to 1C, SAP, ERP, CRM, EDMS, BI, DWH, Service Desk, GitLab, Jira, Confluence and internal API.
RESTART AI Enterprise Platform can be supplied with the RESTART managed AI infrastructure: production/test/dev environments, GPU resources, storage, backup, monitoring and technical support. This allows you to quickly move from pilot to production operation without searching for separate contractors for servers, DevOps and information security.
For the customer, this is a comprehensive offer: platform, computing power, setting up AI agents, downloading a knowledge base, integration with systems, information security requirements and further support.
The platform is confirmed not only by the description of the architecture. On its basis, RESTART is developing applied AI products: 1trAIner for sports and wellness, Spina Bifida Check to support early attention to possible signs on ultrasound, and the RAG foundation assistant with a proven knowledge base.
These projects are different in subject area, but the same in engineering principle: AI works within a controlled process, uses data and sources, has restrictions, logging, roles, integrations and clear human responsibilities.
RESTART AI Enterprise Platform is confirmed by industrial banking implementation: for a bank from the top 5 of Uzbekistan under NDA, RESTART created an AI platform and RAG agents for corporate knowledge, client and internal channels.
The project was piloted, received a positive assessment from the customer and is now being supported. This shows that the platform can live not as a one-time AI experiment, but as a supported enterprise environment with development, quality, access and operation.
RESTART AI Enterprise Platform can be the platform core for a broader product line. At the level of application scenarios, the group is developing CleverHub: a voice assistant, meeting minutes, document processing and a RAG assistant. These products can be displayed as an independent showcase or associated with the RESTART enterprise architecture.
Each module can be supplied as an independent platform extension or as part of a comprehensive enterprise package. The logic of the platform is that the core does not change, and application scenarios are connected to the general rules of security, integration and operation.
A unified environment for enterprise AI agents, models, RAGs, access rights, logs, integrations and auditing.
Agent designer, admin console, role management, logging, auditing, managing models and request costs.Search and answers for corporate documents, regulations, contracts, project documentation and knowledge base.
RAG platform, document loaders, vector storage, source citation, ACLs and permissions.Reducing the workload of the first line of IT, HR and financial support, classifying requests and prompting operators.
Telegram, web, portal, mail, integration with Jira, Naumen, ServiceNow, 1C ITIL and support knowledge base.Plan-fact, management reporting, explanation of deviations, cash-flow, receivables and budget comments.
Financial AI module for 1C:ERP, 1C:UH, SAP, BI, DWH, management and financial reporting.Analysis of contracts, procurement documentation, RFP/RFI, essential conditions, risks and deadlines.
OCR, comparison of versions, extraction of conditions, risk checklist, integration with EDMS, Diadoc and Kontur.Procurement search, relevance scoring, proposal preparation, technical specifications compliance matrix and case selection.
Module for monitoring tenders and preparing sales based on current developments RESTART.Protected AI assistance to developers in a closed loop: code, tests, reviews, documentation and legacy systems.
Integration with GitLab, Jira, Confluence, code search, test generation, review assistant and documentation.AI risk control, GRC, SOC Copilot, Federal Law No. 152-FZ/FSTEC checklists and incident analysis.
SOC/GRC Copilot, incident analysis, information security policies, Federal Law No. 152-FZ audit, reports and compliance checklists.Selection, adaptation, certification, competency base, training and internal HR bot.
HR Copilot, competency profiles, screening, development plans, knowledge base and internal employee bot.Industry-specific AI modules for industry: MRO, shift magazines, HSE, metrology, production and quality.
A set of industry agents on top of the platform, ERP/EAM/MES/ACS TP connectors and industry templates.The architecture is selected according to the customer’s requirements: private cloud, on-prem or hybrid model.
The answers are based on the allowed data; Users shouldn't see anything they don't have access to.
Actions, requests, responses, sources, prompt versions and control events are recorded.
For mission-critical processes, human inspection, quality metrics and gradual commissioning are used.
For scenarios with agent actions and integrations, the platform communicates with protected by AI automation: authority restrictions, logs, source control, human-in-the-loop and operational rules.
We select a task where there is repeatable intellectual work: documents, applications, contracts, tenders, finance, development or information security.
We determine sources, data quality, owners, restrictions, personal data, trade secrets and placement requirements.
We describe the core, integrations, roles, agent scenarios, quality metrics and acceptance criteria.
After the pilot, we record the result, finalize the platform core and connect the following modules.
2-4 weeks: goals, processes, data, risks, architecture and first module.
6-10 weeks: platform core, RAG, roles, logs, integrations and the first working scenario.
Commissioning into limited operation, quality metrics, human review, user training.
Connection of new modules, sources, departments and operating regulations.
The platform is based on several RESTART practices: AI architecture, information security, ERP/1C/SAP, Data/BI/DWH, custom development, DevOps/DevSecOps and dedicated engineering teams. Therefore, an AI project is considered not as the implementation of a separate tool, but as a change in the corporate IT environment.
The RESTART AI Enterprise Platform can be used alongside industrial R&D projects where managed data sources, roles, logs, RAGs for engineering materials, document analysis, operator prompts, and AI connectivity with enterprise systems are needed. In projects like ASB-27, the AI loop does not replace engineering assessment, but helps to work with data, documentation, statuses, reporting and knowledge.
This is a common layer for all the company’s AI services: language models, AI agents, knowledge base, access rights, request log and integration with corporate systems. The platform operates within the customer's environment, and new scenarios are connected to the already created core.
Yes. RESTART AI Enterprise Platform is deployed on-premise, in a private cloud or hybrid. Local language models require a GPU environment; RESTART can deliver it along with the platform as a managed AI infrastructure.
The platform connects models through a layer of models: GigaChat, YandexGPT or open-source models within the customer’s perimeter; the model is selected in the control panel. In the banking case, LLM Qwen3-30B, embedder and reranker are running on the customer’s GPU servers. A specific model is selected for the task and data requirements.
For productivity, the benchmark is from 16 vCPU, 64–128 GB RAM, 1 TB NVMe and GPU level L40S 48 GB or A100 80 GB; The test/staging environment is enough for the pilot. The exact configuration depends on the number of users, frequency of requests and models and is recorded in the technical specifications.
Yes. Each module can be an independent extension of the platform. But even the first module is better designed so that it does not become an isolated experiment.
No. It is usually better to start with one process, prove the effect, configure the core and then scale the platform to new units.
At the discovery stage, data sources, access rights, personal data, logs, placement and human verification scenarios for critical processes are recorded.
For enterprise AI, not only models are important, but also correct data sources. 1C, Bitrix, ERP, document flow, portals and personal accounts often contain key data for AI assistants: contracts, applications, statuses, financial indicators, documents, directories, client requests and regulations. Therefore, RESTART AI Enterprise Platform is designed taking into account ERP integrations, roles, access rights, masking sensitive data and information security requirements.
We will show the product on your environment and offer a pilot format with a clear output result.
