Everything they ask before the contract — and after.

A classic B2B marketplace is optimized for a stream of universal SKUs and listing commissions. RESTOK.PRO is built for industrial supply — where the real difficulty is not the price, but whether a substitution is admissible under GOST, TU and corporate acceptance regulations.

• →The catalog and dictionaries are written for industrial categories: pipeline valves, tubular products, instrumentation, rubber goods, workwear, fasteners, electricals and more.
• →AI analog matching justifies every substitution with a reference to a normative document — never “looks similar by name”.
• →Every deal is backed by escrow, seller verification and arbitration — this is not a classifieds board.
• →There is a corporate perimeter for holdings: internal turnover between subsidiaries without going to the open market.

A good fit: large industrial holdings (oil & gas, metals, chemicals, energy, transport), manufacturing enterprises with a branched subsidiary structure, independent industrial-product dealers, oilfield-service companies.

Not a fit: retail stores, office-supplies sellers, FMCG distributors — dozens of universal venues will serve them better and cheaper.

If your typical order runs into millions of rubles and the deciding question is “does it pass the TU”, you are in the right place.

The team currently works with 10+ industrial assets — subsidiaries and manufacturing enterprises. Priority industries:

• →Oil & gas and oilfield services — historically the founder's core expertise.
• →Metals and mining — corporate perimeter for groups of enterprises.
• →Chemicals and petrochemicals — analog selection by GOST and acceptance regulations.
• →Energy and power generation — work with instrumentation and electrical reference books.

Machine building and transport are on the expansion roadmap.

The base normalized reference catalog covers hundreds of thousands of industrial items across 18 categories. Every card is built on a triplet: normative document (GOST / TU / OST / DIN / ISO) + parametric profile + corporate part number.

When a client connects, their own reference catalog is matched to ours within 3–7 business days — and from that moment on, every one of the client's items gets AI analog matching.

The web interface runs in modern browsers: Yandex Browser, Chrome, Firefox, Safari. For mobile access there is a responsive version and a dedicated app for executives and field buyers (iOS / Android).

AI search, request creation, deal signing and report export are all available in the mobile version as well.

Yes. The Start plan is free: create an account, list a basic catalog and test AI analog matching and the deal flow with no subscription fee.

For holdings there is a dedicated pilot: we deploy one perimeter with two subsidiaries for 60–90 days and decide on scaling based on the results.

A marketplace earns on traffic — listings, promotion, paid product cards. A platform earns on protecting the deal — escrow, verification, insurance and arbitration.

We have no paid card promotion and no “VIP lots at the top of the list”. Search is always objective and driven by normative parameters — not by who paid more.

The platform was created by KorpService LLC — a company built around a founder with 20+ years in leadership roles in materials management for large industrial assets. The idea was formulated in 2023, engineering began in 2025, and in 2026 the platform is in pilot deployment with real counterparties.

The full timeline and the team are described on the About page.

The platform runs on a pair of models:

• →The senior model — an open language architecture of the latest generation working together with the platform's industrial corpus: catalogs, GOST/TU standards, engineering dictionaries and labeled “item — analog” pairs. Analog compatibility is checked by deterministic rules, not by the neural network's “opinion”.
• →The lightweight model handles simple tasks (search, classification, clarifications). Faster and cheaper, with no quality loss on routine operations.

By default, processing runs on servers inside the Russian perimeter (local models, bge-m3 embeddings). Anonymized queries are sent to foreign AI providers (GLM, China; Claude, USA) only with the user's explicit consent — it is disabled by default. See section 4 of the Privacy Policy.

No. A client's queries and responses are visible only to their own employees and never enter the training set. We enforce strict customer isolation — there is no cross-training between counterparties.

General model improvements are made on public normative sources (GOST, TU, open catalogs) and on the team's own labeled corpus. Client data never gets there under any circumstances.

All servers running the platform — catalog storage, vector representations, query logs, deal documents — are physically located in data centers in the Russian Federation, certified to the requirements of laws 152-FZ and 187-FZ (critical information infrastructure).

By default there is no cross-border transfer — processing stays within the Russian perimeter. Anonymized queries can be sent to foreign AI only with the user's explicit consent (disabled by default), and even then the name, email, phone and company details are never transmitted. See section 4 of the Privacy Policy.

We measure matching quality on an internal set of real industrial queries: every suggested analog passes deterministic engineering checks (diameter, pressure, connection, standard family). Accuracy comes from industry labeling — real “item — analog” pairs, flagged incorrect substitutions and encoded GOST/TU rules.

We deliberately do not publish an averaged “accuracy percentage”: it depends heavily on the nomenclature and category. We demonstrate concrete metrics on your own catalog during the pilot.

No — and we deliberately do not build it that way. Fully automatic analog substitution without a human is a direct risk of shutting down a facility: one wrong valve substitution at high pressure can cost more than a year of automation savings.

Our AI does three things:

• →Proposes options with justification against the normative document.
• →Highlights substitution risks — “different pressure class”, “different seal material”, “non-standard design”.
• →Leaves the final decision to the procurement specialist.

Both. Structured-field search fires first — part number, manufacturer, normative document. If there is no exact match, semantic search over the parametric profile kicks in: material, connection type, nominal diameter, pressure class, climatic rating and so on.

Results are ranked on a combination of factors: parametric proximity, availability from verified sellers, delivery time, and the client's corporate constraints.

Every answer comes with a justification and the model's confidence. On questionable substitutions the system explicitly declines to recommend and suggests escalating to a specialist.

If a specialist flags a match as wrong, it enters the feedback loop — not to train on your data, but for manual review by our industry experts and an update of the shared substitution rules.

The commission is 8% of the deal amount. The buyer pays it on top: the seller receives 100% of the goods' value. Intra-holding transfers between subsidiaries are 0% (only logistics is paid).

What the commission covers: legal protection of the deal, arbitration, the support desk, counterparty verification, and the platform's AI services.

No. Escrow, deal insurance and factoring are separate costs paid directly to partner banks and insurers.

• →Escrow — at the bank's tariff; fixed in the deal estimate before signing.
• →Insurance — at the insurer's tariff (depends on cargo category and route); arranged when booking delivery.
• →Factoring — a partner bank's service for the seller; the rate is set individually by the bank and is unrelated to the platform's commission.

We show the full deal estimate in a single window before signing — no hidden fees.

Escrow is conditional holding of the buyer's money until the seller fulfils their obligations. The flow:

• →The buyer transfers the deal amount to the partner bank's escrow account.
• →The seller receives confirmation and ships the goods.
• →The buyer accepts the goods against an acceptance act (or files a reasoned claim).
• →The bank automatically releases the escrow to the seller after acceptance.

In a dispute, the deal moves to the arbitration module — documents, correspondence and acceptance parameters are already inside the platform, so nothing has to be collected manually.

The 8% commission is paid by the buyer — on top of the goods' value. The seller receives 100% of the offer price. The product card shows the seller's price; the commission is calculated on top and shown as a separate line in the deal estimate before signing.

Intra-holding transfers between subsidiaries carry no platform commission (0%).

The base commission is flat — 8% of the deal amount, the same for everyone: we have no volume-dependent rates. For large corporate clients, individual commercial terms are discussed within a framework agreement.

If the buyer declines to sign the acceptance act that releases the escrow, the arbitration module starts automatically. It follows the procedure set out in the terms of service:

• →First 3 business days — negotiations between the parties inside the deal chat.
• →If no agreement is reached — escalation to the platform's arbitrator with the full document package.
• →The arbitrator's decision is the basis for a partial or full escrow refund.

For a justified return, the commission is refunded to the buyer. The escrow and the insurance premium are refunded under the bank's and the insurer's procedures.

All prices on the platform are shown inclusive of VAT. The platform's commission includes VAT. Invoices and universal transfer documents (UPD) are generated automatically, with VAT itemized at the seller's rate (typically 20%).

Companies on the simplified tax regime (USN) must explicitly declare it on the seller card — the platform then issues documents without VAT.

Holding mode is a closed corporate perimeter inside the platform, where several subsidiaries of one group see each other's internal stock, needs and surplus — without going to the open market.

The parent company gets a consolidated report across the whole group: where there are shortages, where there is surplus, and where internal turnover can replace external procurement. In effect, it is the group's built-in supply logbook.

Technically there is no limit: the holding perimeter is designed for groups of 2 to 50+ subsidiaries. A typical pilot covers one or two subsidiaries for 60–90 days before scaling. Pricing depends on the number of users and catalog volume, not on the number of entities.

By default — only the two parties to the deal and authorized head-office employees with the appropriate role. Visibility is flexible: a deal can be fully private between two subsidiaries, or transparent to the entire perimeter.

Nobody outside the holding ever sees internal deals. Ever.

Yes — that is the standard scenario. The system first looks for the item within the group's stock. If it isn't there — or the terms are unfavourable — the request goes to verified external suppliers. The policy is configurable: “internal first, then external”, or “parallel search”.

Subsidiaries' illiquid stock automatically enters the corporate surplus showcase. Other group companies see it first — before it reaches the open market. That cuts from-scratch procurement where the item is already sitting in another group warehouse.

If internal turnover does not happen within a set period (usually 60–90 days), the item is published to the external showcase — now flagged “approved for sale by the parent company”.

A typical roadmap:

• →Weeks 1–2 — export and matching of the client's reference catalog to our normalized catalog.
• →Weeks 2–3 — configuring users, roles, corporate rules and limits.
• →Weeks 3–4 — integration with the accounting system (1C, SAP, in-house systems), test deals.
• →From week 5 — production operation.

For holdings, the first wave — one or two subsidiaries — takes 4–6 weeks. After that, each new entity connects in 1–2 weeks.

Yes. There are ready-made connectors for 1C:ERP, 1C:UPP and 1C:Holding Management. Standard scenarios synchronize:

• →Reference books of items, counterparties and contracts.
• →Needs and purchase requests.
• →Deal documents — invoice, UPD, acceptance act.
• →Warehouse stock and item statuses (surplus / main / reserve).

SAP has a working integration via an exchange bus (MQ or REST, your choice). Oracle E-Business Suite and in-house corporate systems connect through the platform API (REST/JSON, documentation is open). For a non-standard stack, we build an adapter in 2–3 weeks of our engineers' time.

The platform is fully self-hosted and runs within the Russian perimeter — no dependence on foreign clouds. For clients with elevated requirements we discuss individually:

• →Private cloud — a dedicated perimeter in our secured data center.
• →Dedicated perimeter — a separate platform installation for the client with scheduled updates.
• →Hybrid schemes — the scope and boundaries of perimeters are defined during pilot design.

Dedicated-perimeter options are priced individually — contact the sales team.

No. The platform is configured to the client's existing regulations: approval thresholds, mandatory tenders, single-purchase limits, preferred supplier lists, acceptance rules. All of these are parameters, not hard-coded logic.

If the rollout reveals that a regulation is outdated, we help draft the updates — but never force them. That is the customer's decision.

During the rollout — we do. The standard package: an introductory workshop for executives (2 hours), key-user training (1 day on-site or 2 days online), administrator training (2 days).

On top of that — a knowledge base, training videos and regular webinars on new features. Enterprise clients get a dedicated trainer for the first 3 months.

Yes. The platform:

• →Processes personal data within the Russian Federation (152-FZ); the data operator is KorpService Consulting LLC (registration details are in section 9 of the Privacy Policy).
• →Is built with the protection requirements for critical information infrastructure in mind (187-FZ).
• →Processes data in the Russian AI perimeter by default; cross-border transfer of anonymized queries happens only with the user's explicit consent.

Processing details and the list of engaged processors are in the Privacy Policy (sections 3–4, 7, 9).

All major Russian e-document operators are supported: Diadoc, SBIS, Kontur, Taxcom and others — via roaming. Deals are signed with a qualified electronic signature (QES) using CryptoPro or VipNet CSP.

For closed corporate perimeters, integration with the client's internal EDI system is possible.

The parties to the sale contract are always the seller and the buyer. The platform is not a party to the deal and never takes ownership of the goods.

The platform acts as the deal facilitator and the escrow settlement operator, which is fixed in a separate agency agreement. This matters for correct tax and bookkeeping treatment on the client's side.

RESTOK.PRO is not a trading venue under laws 44-FZ or 223-FZ; purchases made through the platform are not regulated procurement. The platform is designed for intra-corporate turnover and purchases on the open commercial market.

For companies operating under 223-FZ there is a separate scenario: analog selection and verification on the platform, followed by formalizing the deal through the regulated procedure on the customer's trading venue.

An NDA and a trade-secret agreement are signed with every client. Inside the platform itself:

• →Deal content and correspondence are isolated between counterparties.
• →Access logs are kept per user, with detailed reporting for the client's security service.
• →Data marked as a trade secret never enters the platform's aggregate analytics.

Response times are fixed in the contract SLA and confirmed by a monthly report.

On the Business and Holding plans — yes. A dedicated manager accompanies the client from rollout to production operation, tracks utilization, suggests optimizations and collects feedback on new platform features.

The availability target (contractual SLA) is 99.9% during working hours (8:00–22:00 MSK, Mon–Sun), with planned maintenance windows at night.

For the Holding plan, the target level and response times are fixed in the contract SLA — see the table above.

The platform ships continuously: minor releases weekly, major feature blocks quarterly. A six-month roadmap is available to clients in their dashboard, with voting on priorities.

Every change is published in the release log with a date, a description and instructions — nothing changes silently.