Ney asks your AI employees, reports in Inbox, Ney inside your CRM

What's new
ChatNeuron

trust center

What we do with your data — and what is still in progress

Everything a security team checks: where servers are, who can do what through the assistant, what the model sees, which documents exist and which do not yet.

  • 01Servers in KazanIT Park Kazan data centre; ChatNeuron is an IT Park resident.
  • 02X-ray reads, actions on “Apply”Diagnostics only read. The assistant sees what the employee sees and by default changes nothing without confirmation.
  • 03Audit logWho, when, which action, how it ended — without chat texts.
  • 04Honest statusesRegistries and checks we do not have yet are marked “in progress”.

reading and actions

Diagnostics reads, actions need your consent

Diagnostics (X-ray) — read-only

The X-ray reads deals, calls, payments and statements and shows specific problems: deals with no task, calls with no callback, overdue payments. It creates, changes and sends nothing in your systems.

Assistant actions — after your confirmation

  • By default the mode is “Ask first”: anything that changes data or goes outside is shown as a card and done only after you click “Apply”.
  • You turn on “Autonomous” mode yourself: then the assistant does safe actions on its own, shows “Done” with an “Undo” button and writes them to the log.
  • Money and deletion always need confirmation, in any mode. Team permissions, access, outgoing letters and turning on customer replies are always a card too.

AI employees (customer replies, requests to the CRM, bookings) act by the rules you switch on in their settings.

access

Who can do what through the assistant

The assistant does not get a master key. Every request comes from a specific person, and your system decides what to show them.

  • 01

    On behalf of the employee

    The system is told who is asking. No permission for a section — the assistant says so instead of working around it.

  • 02

    Reads at once, changes by button

    By default creating a task, sending an email or adding a record is only proposed: you see a “what will change” card and press “Apply”. You turn on “Autonomous” mode yourself — then safe actions run on their own, with “Undo” and a log entry; outgoing letters, money and deletion are still a card.

  • 03

    Re-checked on apply

    Permissions are checked again between the proposal and “Apply”: revoke access and the action will not run.

  • 04

    Audit log

    Every call to a system is recorded: who, when, which action, how it ended. No chat texts or system responses.

  • 05

    You enable actions yourself

    When you connect your own system, reads are on and anything that changes data stays off until the owner enables it.

  • 06

    Access is revoked instantly

    The key is stored encrypted and never displayed. Revoke it on your side and access is gone.

masking

The model does not see personal data

Before text is sent to the model, personal data is replaced with tokens. In the answer the tokens are swapped back — you see normal text the model never saw.

what the customer wrote

My name is Anna Smirnova, phone +7 912 345-67-89, email anna@romashka.ru. Passport 4510 123456.

what the model received

My name is {name_1}, phone {phone_1}, email {email_1}. Passport {passport_1}.

  • phone numbers
  • email
  • full names
  • passport
  • tax ID (INN)
  • insurance number (SNILS)
  • date of birth
  • driving licence
  • address: house and flat

Masking is on by default and works in customer chat, in the assistant inside your systems and in the dashboard helper. In work systems the names of colleagues and customers are kept — without them the assistant cannot name the person responsible.

models

Which models run and where data goes

A separate page: Russian and foreign models, what the model receives, the “Russian models only” switch, your own model and the box.

Models and data
  • Applies to everyone

    Customer-facing AI employees, assistants inside your systems and the dashboard helper.

  • No silent fallback

    If the Russian model is unavailable, the assistant refuses. The chat is not sent to a foreign model.

  • When to enable

    When the assistant handles personal data of customers or staff and it must stay in Russia.

where it runs

Cloud, cloud with your own model, or on-premise

Your own model in the cloud gives you control over AI: no third-party AI provider sees your questions or documents. Storage still stays with ChatNeuron. Keeping all data in-house is only possible with the on-premise edition.

Discuss on-premise

Cloud

Where AI answers
ChatNeuron cloud models (or Russian-only)
Who sees the data
ChatNeuron and the cloud model and search provider
Where data is stored
ChatNeuron servers

Cloud + own model

Where AI answers
A model on your company's server
Who sees the data
ChatNeuron. No third-party AI provider, once both your model and your own embeddings for search are connected
Where data is stored
ChatNeuron servers — nothing is moved when you connect

On-premise

Where AI answers
A model on your servers
Who sees the data
Only your company
Where data is stored
Your servers

Connect your own model in the dashboard: AI models → Company model — a model server and an embeddings server (for example, Ollama with bge-m3). We install the on-premise edition on your servers for evaluation.

storage

What we store and what we do not

We do not copy your database. The assistant requests only what is needed to answer a specific question.

We store

  • chats with the assistant — like chat history
  • the action log: who, when, what was called
  • access keys — encrypted only
  • the knowledge base you uploaded yourself

We do not store

  • a copy of your CRM, mail or accounting system
  • your systems’ responses in the action log
  • keys in plain text — even your admin cannot see them
  • action parameters with personal data: they are masked in the log

honestly

Where the protection ends

We do not promise the impossible. Here is what to know before connecting.

  • Masking is not magic

    It recognises data by format. A rare surname without context, a company name or a deal amount may reach the model as is.

  • Use the Russian mode for personal data

    If chats contain personal data, enable “Russian models only”. Masking then becomes a second line of defence, not the only one.

  • System data is not a command

    The assistant treats the text of an email or a request as data. An instruction inside an email is not executed, and any action still needs “Apply”.

152-fz

Working under the personal data law

You remain the operator of your data; we process it on your instruction and only to run the assistant.

Privacy policy
  • Servers in Kazan IT Park

    Data is stored in the Kazan IT Park data center with 6-level protection. ChatNeuron is an IT Park resident.

  • Documents

    Public offer, personal data policy, contract, invoices and acts for companies.

  • Company data is separated

    One account’s data is not available to another. Dashboard access is role-based: owner, admin, editor, operator, observer.

  • Russian models

    In “Russian models only” mode chats are processed by Russian models; foreign ones are not used for answers.

documents and registries

What exists and what does not yet

We do not show numbers and certificates we have not received. Each row is the real status on the date below.

  • Available

    Personal data processing policy

    Published, with the list of providers and the processing location (Russian).

  • Available

    Personal data consent and marketing consent

    Separate documents, separate checkboxes in every form (Russian).

  • Available

    Offer, terms, company details

    The operator is a sole proprietor; a move to an LLC is being prepared (Russian).

  • In progress

    Entry in the Roskomnadzor personal-data operators registry

    No registration number yet; it will be shown once the entry is confirmed.

  • Planned

    Russian software registry (Ministry of Digital Development)

    No application filed: we are first determining the rights holder.

  • Planned

    Independent penetration test

    No results yet, so we make no claims.

  • Planned

    Certificates and attestation

    We have no certificates yet.

  • In progress

    Status page with incident history

    Current service state is on the Status page; there is no incident history yet. Support is the website chat and the in-dashboard assistant.

Statuses as of 05.11.2026.

for security teams

Document package for review

Leave a contact — we will send what is ready today and tell you what is still being prepared. The request goes to our CRM with your consent to data processing.

Package contents

  • Personal data processing policyAvailableAlready on the site; we will send the current edition.
  • Architecture and data-flow descriptionIn progressDraft ready, under review. What goes where in the cloud, with your own model and in the box.
  • Protection measures under Federal Law 152In progressDraft ready, under review. Access, key encryption, the log, separation of account data.
  • Data-processing agreement templateIn progressDraft with our lawyer. You are the operator of your customers’ data; we process it on your instruction.
  • NDA templateIn progressDraft with our lawyer; prepared for your company on request.

The closed perimeter and box have a separate description for security on the page For large companies

Request the package

We need a company and a contact: email or phone.

See it on your own data

Connect a test system and open the log: what the assistant requested, on whose behalf, and what reached the model.