Trust
NiuCore uses AI models from external providers. Instead of claiming that information never leaves the platform, this page sets out what is sent, in what form, and what remains under NiuCore's control.
The message reaches NiuCore over an encrypted connection.
The area, libraries, connectors and models enabled for that person.
The three layers replace sensitive data with placeholders. The link between placeholder and data stays in NiuCore.
The model receives the text with placeholders and returns its response.
NiuCore restores the data and the employee receives the complete response.
Consumption traceability is kept: user, date, model and niucredits.
| Information | Location | Note |
|---|---|---|
| Conversations and configuration | NiuCore's infrastructure on Amazon Web Services and Microsoft Azure, providers certified under ISO/IEC 27001 | Isolated by company. Information is encrypted at rest and in transit |
| Documents uploaded to a library | The original file is not kept: it is processed and deleted. NiuCore keeps the text extracted from the document in its storage on AWS, within its own infrastructure | That text is not sent to model providers: the search index and everything the model receives carry sensitive data already replaced. Accessible according to the area's permissions |
| Document search index | NiuCore's vector database | With anonymization active, the text is indexed already replaced |
| Dictionary and placeholder–data link | NiuCore's infrastructure | Not sent to the model provider |
| Files in Drive, OneDrive and other connectors | In the company's account with the original service | From the conversation, NiuCore accesses the original document directly, with each user's authorization, and does not copy it. Only if the document is added to a library is its content indexed |
| Payment data | Mercado Pago | NiuCore does not receive the card number |
NiuCore provides access to models from OpenAI, Anthropic, Google, xAI, DeepSeek, Alibaba (Qwen), Moonshot, Zhipu and ByteDance. All of them receive the same treatment: text with placeholders instead of sensitive data.
Profesional Standar and Profesional MAX licenses give access to every model. If company policy requires it, each area can be restricted to approved providers; a private area can operate exclusively with them.
Anonymization happens before the provider is selected. No provider receives unreplaced data.
Besides the model that answers, NiuCore uses models to select the model in Auto mode, summarize long conversations and generate next best action recommendations. These tasks follow the same anonymization flow and run on the best available model among those enabled in the user's area; they therefore respect each area's provider restrictions. Library indexing uses OpenAI models and receives text that is already anonymized.
In Auto mode, NiuCore switches to another of the models enabled in the area so work can continue. With a fixed model, it reports that the model is unavailable instead of replacing it.
NiuCore does not train its own models with its clients' information. It accesses third-party models through their enterprise services or their application programming interfaces, not through their free or consumer versions, and contracts under terms that exclude the use of information for model training with every provider whose standard API terms provide for it. The table below states each provider's position; if your organization's policy requires it, areas can be restricted to the providers that exclude it.
NiuCore contracts with each provider, wherever its standard API terms provide for it, under terms that exclude the use of information for model training or for any other unauthorized purpose. Even so, neither NiuCore nor your organization can verify from the outside that a third party keeps its promises. That is why active anonymization is the core of the platform: replacement happens inside NiuCore, before anything is sent. Names, identity documents, contact details, contracts and your organization's own terms travel as placeholders —⟦Person#7F3A2C⟧, ⟦Document#B81E44⟧, ⟦Contract#4E91D0⟧— and the mapping between each placeholder and the real data stays within NiuCore: no provider receives it. A provider that retains the text it receives, or uses it to improve its models, retains placeholders, not your organization's data.
All of them receive text with placeholders instead of sensitive data. This table also sets out what each provider's terms say about the information it receives.
| Provider | Use of information for model training | Where it is processed | Retention by the provider |
|---|---|---|---|
| OpenAI | Excluded under its API terms | United States | Up to 30 days, for abuse detection |
| Anthropic | Excluded by contract | United States | As set out in its commercial terms |
| Excluded in the paid service NiuCore uses | Countries where Google maintains facilities | A limited period, for abuse detection | |
| xAI | Excluded unless expressly authorized | The provider's global infrastructure | 30 days, for abuse detection |
| Alibaba (Qwen) | Excluded | Singapore | No period published |
| Zhipu | Excluded unless expressly agreed | Singapore | The provider states that it does not store content |
| ByteDance | Excluded unless expressly authorized | Malaysia, Indonesia or Europe (EU/EEA) | Content that triggers its safety filter is retained for 180 days, in Malaysia |
| Moonshot | The provider reserves the right to use information to improve its services | Singapore | No period published |
| DeepSeek | Its terms do not exclude it | China | No period published |
Information taken from each provider's public terms, reviewed in [MONTH AND YEAR OF PUBLICATION]. NiuCore updates this table when a provider changes its terms.
Employees' messages and conversation history, library documents, attachments, images and scanned documents, information delivered by connectors, and next best action recommendations.
Placeholders are restored so the employee receives the response with the real data.
No detector recognizes every piece of sensitive data.
It does not intercept other applications.
Each company's information is kept separate. No content—including what a company marks as public—is visible to another.
Each user accesses only the libraries, connectors and models of their areas. Permissions are defined by capabilities, not by job titles.
Company administrators with the corresponding permission can view their users' conversations and contexts, including those the user keeps private from colleagues. No other company has access to them.
Available for user accounts and enforceable by the company.
The company defines which actions it authorizes on each connector: for example, allowing emails to be read but not sent.
Operational memory belongs to each area, is not combined with that of others, and is stored with sensitive data already replaced. Each user sees only their own recommendations, and the administrator can delete an area's memory.
Operational continuity
If an AI provider becomes unavailable, and in Auto mode, NiuCore switches to an alternative model. If a conversation is interrupted, the platform reconciles it and keeps what was generated. The Terms and Conditions set out a 99% monthly availability commitment, its calculation formula and the advance notice for maintenance windows.
The processing of personal data is governed by NiuCore's Privacy Policy and by Colombian personal data protection law, in particular Law 1581 of 2012. With respect to the information your organization uploads to or generates in the platform, your organization is the Data Controller and NiuCore acts as Data Processor, on its behalf and in accordance with the data transmission annex to the Terms and Conditions. Data subjects may exercise their rights of access, update, rectification and deletion by writing to privacy@niucore.com; inquiries are answered within a maximum of ten business days and claims within a maximum of fifteen. When the contractual relationship ends, NiuCore deletes the hosted information three months after the effective termination. NiuCore Technology S.A.S., NIT 901.967.332-9, Colombia.
The documentation describes how the platform and active anonymization work, with the level of detail a formal evaluation requires.
Inquiries from IT and legal teams are answered in writing through Contact.
