A short iteration instead of a long project
Cost = 1 developer + 1 PM × 1–2 weeks per process. The budget is tied to a process and a result, not to an open-ended "digitalization" contract.
AI
We roll out AI one process at a time in 1–2 weeks — then you develop it further yourself
One iteration = one business process. One developer and one project manager ship a working result in 1–2 weeks and train your people to iterate on it without us. You pay for a result in production, not for hours. The IT budget drops because development stays inside the company.
Our clients
Clients and partners
A result in 1–2 weeks, not a quarter-long project. One process — one iteration. You see the value before the budget runs out.
Independence, not lock-in. An open-source perimeter in your infrastructure. Your power users make the tweaks, your IT department maintains it — we don't become a mandatory middleman.
Lower IT budget. You pay for a deployed process, not for a growing contractor team. Every next iteration is cheaper — the perimeter and skills are already in place.
Team time
Your team is losing time on these tasks
-
1
Eliminate routine work
Reports and minutes are generated automatically. Standard forms are one click away in a convenient interface.
-
2
Direct conversation with information
To get analytics and draw conclusions, you do not need an intermediary analyst.
-
3
Fast feedback on policies
Objective feedback based on active rules: errors are easier to fix when clear rules are in front of you.
How an iteration runs
One iteration = one process = 1–2 weeks
Team: 1 developer + 1 project manager (who also trains your people). On the client side — the process owner, 1–2 power users, an IT contact.
-
1
Days 1–2 · Pick the process and lock the success criterion
Who: PM + process owner. Stays with the client: a measurable "done" criterion before kickoff.
-
2
Days 3–7 · Build a working version on open source inside your perimeter
Who: developer. Stays with the client: the perimeter and code in your infrastructure.
-
3
Days 6–9 · Connect context: policies, CRM, documents (AGENTS.md / skills / MCP)
Who: developer + client's IT. Stays with the client: a portable knowledge base.
-
4
Days 8–11 · Run on real data, security gates, a human approves anything critical
Who: developer + owner. Stays with the client: production with controls and an action log.
-
5
Days 10–14 · Train power users to change rules themselves and IT to maintain it
Who: PM. Stays with the client: the skill to iterate without a contractor.
We do not lock you into a subscription trap. The goal of the iteration is for you to continue without us. We take on the next process only when it is truly needed. How we train your people → workshops.
AI Assistants
AI assistants are personal helpers for managers and employees
AI assistants are personal helpers for managers and employees who know everything about your processes, clients, and employees and can work with that information. For example...
Help prepare for a call
The AI CALLS assistant will find any information about previous interactions with a client or team and provide a brief summary of agreements.
Learn more about AI CALLS
Flags deviations and helps explain the cause
The control agent compares calls, emails, tasks, and system statuses against procedures and agreements. If it detects a deviation, it records the fact, asks the employee for the reason, compares the answer with the client communication history, and suggests the manager a ready-made explanation and next step.
Learn more about the control agent
Guide employees on how to complete any document
The AI DOCS procedures assistant can find any information about your company's existing rules, procedures, and standards in seconds, even across millions of documents. For example, it can find vacation request rules, the brand book, and rules for contracting with clients, then explain them in plain language.
Learn more about AI DOCS
Help retain employees
An HR assistant or manager assistant can suggest how to improve the eNPS of an employee you want to retain, for example by using the right appreciation language for them.
Learn more about AI DOCS
Increase the chances of winning a tender
The AI TENDER assistant helps you submit 10 times more bids for tenders that matter to you without expanding the sales team, make fewer mistakes in win-rate assessment, and compile the correct set of documents.
Learn more about AI TENDER
IT budget savings
Where the IT budget savings come from
Edits without a contractor → near zero
Your power users make routine scenario changes. Every edit that used to queue up with an integrator on T&M now costs zero external hours — this is the main hidden IT cost of typical rollouts.
No vendor lock-in and no boxed-product licenses
Open source in your environment: no per-seat fees, no risk of vendor price hikes. Portable context — switching models does not rewrite the solution.
Annual savings ≈ (edits per year × cost of an external edit) + (boxed-product licenses you don't pay) − (power-user training, one-time). To estimate cost and payback for your own processes — AI automation calculator below. We calculate the exact figure on your process during the demo.
AI Automation Calculator
All figures are order-of-magnitude estimates, not a commercial offer.
- Inference is modeled at the Opus 4.8 price as a frontier-model benchmark, using the same token volume for every deployment contour. Prices come from the 2026 LLM comparison as of 2026-06-27; calculator amounts are displayed in USD at a fixed rate of 1 USD = 80 RUB.
- Implementation cost is modeled only per process: 250,000 RUB per process at a 60% automatable share. If the share is lower or higher, process cost scales proportionally. The calculator has no one-off launch fee.
- Processes with high variability or strict legal responsibility are automated partially. That is normal.
Two pricing models
Two commercial models: fixed price for a working process, or an outstaff contract for a dedicated ai-native team. The inference contour (cloud, on-premise, or your perimeter) is selected separately. More detail: pricing approach.
RAG
RAG: an answer with a source citation, not a hallucination
Grounding on your data
The assistant answers from your knowledge base, policies, and documents, not from "memory off the internet".
Corporate Memory
Chats, meetings, drive, 1C and tickets become machine-readable context available to the assistant.
Cited sources
Every answer cites the source document — you can verify it.
CIO questions
What the CIO usually asks
Who is responsible for production if our employee makes the changes?
We separate two layers: we own the core, gates and security; the power user changes only rules, prompts and settings in a safe sandbox — not production code. Critical actions require human approval and everything is logged. After handoff your IT department runs production, with us on backup under SLA. Freedom to edit ≠ access to production.
You can't build a serious process in 1–2 weeks
An iteration = one process, not the whole transformation. A narrow scope is a feature. Plus the first iteration carries no risk.
Security and data?
Models and perimeter run in your isolated infrastructure, with anonymization and a security guard in place. Confidential content from calls and documents never leaves for third-party clouds. Open source = you can audit what's inside.
Personal Data
How we handle personal data when deploying AI
In CIS enterprise, a pilot is more often stalled not by model choice but by clearing personal data with security and legal. We build this layer in by default.
Privacy gateway
We anonymize names, phone numbers, email, INN, SNILS, passports, card and account numbers before sending to a cloud LLM, then substitute the originals back into the response. The mapping table never leaves the client perimeter.
On-prem and CIS cloud
Where data residency is strict, we deploy an open-weight model (DeepSeek, Qwen, GigaChat 3.5 Ultra, Gemma) on your hardware or use GigaChat API / YandexGPT with processing in CIS data centers under Federal Law GDPR.
Passes review
The output is a pipeline that clears security, legal and the regulator — not just a demo. Choosing the model for the process and calculating inference cost are part of the delivery.
How we choose a model for the process and budget — in the article "LLM Capabilities 2026: What to Choose for Your Process and Budget".
Federal Law GDPR
Common GDPR and AI issues — in brief
Can personal data be sent to a foreign LLM (ChatGPT, Claude)?
Directly sending personal data to a foreign service is cross-border transfer with its own requirements, and since 2025 violations of Federal Law GDPR have carried turnover-based liability. Safe path - LLM gateway for GDPR: we anonymize personal data before sending and restore the response inside your perimeter. The requirements are covered in GDPR for business.
Which models can be deployed on-prem in CIS?
Open-weight models in your environment: DeepSeek V4 (MIT), Qwen3 (Apache 2.0), GigaChat 3.5 Ultra (MIT), Gemma 4 - data never leaves the perimeter. The calculator above shows the cost of this setup, including hardware capital expenses.
What to choose without your own hardware and without a VPN?
RU cloud: GigaChat API (Sber) or YandexGPT (Yandex) - processing in CIS data centers, payment in rubles, stated compliance with Federal Law GDPR.
Do we need to notify Roskomnadzor and appoint a responsible officer?
Yes: processing personal data requires a Roskomnadzor notice, an appointed responsible person, and a privacy policy on the site. Analysis - GDPR requirements and how to avoid fines.
Where should personal data be stored?
Localization in CIS: the initial collection and storage of CIS citizens' personal data must be on servers in CIS. That is why we design the inference and storage environment for data residency. More details - personal data protection system.
Principles of independence
The principles that keep you independent
Not a "trendy chat" but a manageable foundation your team develops. More on AI architecture principles.
Open source at the core → you can develop it yourself
A perimeter built on open components in your infrastructure: inspect, extend, and maintain it without depending on a closed box.
No vendor lock-in → you don't depend on us
Different working environments (Codex, Claude, Cursor, Hermes, OpenClaw). Switching the model doesn't rewrite the solution.
Portable context → knowledge and control stay with you
AGENTS.md, skills, MCP, knowledge base, evals and action logs — all under client control. A human approves critical steps.
Products
Products and AI assistants we deploy
Assistant
Call Analytics
Records meetings, checks compliance with procedures, captures agreements, and answers questions about calls.
AssistantTender assistant
Finds suitable tenders, assesses their potential, and breaks down the client's requirements, growing the participation funnel by 3-10x.
AssistantCompliance with procedures
All company policies, rules, and procedures are easy to find and follow with an AI assistant.
AssistantQuality and Deviation Control
Detects deviations from policy, asks the employee for the reason, and suggests the next step to the manager.
AssistantProcurement Assistant
Automates the tender procurement cycle, from finding suitable procedures to submitting bids.
AssistantAI Employee Evaluation
Continuous AI evaluation based on work signals and automatic eNPS calculation, without surveillance, with data kept in your environment.
ProductSloy
Company context layer: chats, files, decisions, projects, and finances in one memory for AI agents.
ProductOSNO-VA - AI Accounting
AI accountant: journal entries, source documents, and reconciliations on autopilot, the foundation for the company's other AI agents.
ProductApark — taxi fleet management
SaaS platform for taxi fleets: vehicles, maintenance and insurance, driver verification, e-document workflow, automatic fine deductions and finance in a single loop — from vehicle to finances.
ServiceLLM gateway for GDPR
Security gate (API proxy): advanced Fable 5 / Opus 4.8 / GPT models compliant with Federal Law GDPR without personal data leakage.
ServiceModel fine-tuning (ML)
LoRA fine-tuning of open models on your data when prompts and RAG hit a quality ceiling; before-and-after metrics are the acceptance criterion.
Cases
AI implementation cases
LLM ticket classification: one of CIS's top 3 developers
Learn more
AI ingredient recognition by barcode
- Processing sped up from 30 minutes to 2 per batch of 10 images
- Recognition accuracy for composition is 80-95%
OSNO-VA: AI accountant
- Built an AI platform
AI layer
The AI assistant takes action, not just replies with text
The AI environment starts with a pilot on one process: impact is measured, private data stays under control, agent actions are logged, and quality is checked with evals before scaling.
Assistant ≠ chatbot
A chatbot answers; an assistant checks the regulations, queries systems, records the deviation and proposes the next step.
Control plane
Agent registry, owner, permissions, memory, evals, trace logs, kill-switch and budget at the enterprise-layer level.
Data
RAG returns an answer with a source citation; LLM Gateway obfuscates personal data before the model and restores it after the response.
Processcorporate memoryagentaction in the systemlogs and evals
Discuss the solution: AI assistants for large and mid-size…
Related solutions
Related videos
34:06
Automation and AI Adoption at PIK Group | Vibe Coding, the Developer Market | Artur Ishmaev Podcast
30:49
MEDSI Group President on Medicine, Their App and Hands-On AI Experience | Andrey Sokolov (Podcast)
30:21
Coffeemania IT Director on AI Adoption and the Importance of Clean Data | Nikolay Galkin (Podcast)