Your Data and Model Performance
What happens to your content, and how we keep AI models accountable
Last updated: 3rd August 2026
This page explains, step by step, what happens to your content when you use an AI-assisted feature on NSOffice.AI, and how we keep the models behind those features accountable for quality and safety over time.
1. We Do Not Train Models on Your Data
Your organization's chat content, documents, and connected business data are never used to train NSOffice.AI's own models or any third-party AI provider's general-purpose models. Your data stays your data — it improves your own results within your own organization, not a model shared across our customer base.
2. How a Request Is Processed
- You submit a request - a chat message, an uploaded document, a query routed through a connected business system, or a specialized AI agent ("Bee").
- Our guardrails layer screens the request first: detecting and redacting personal data, and checking for unsafe or manipulative content.
- The screened content is sent to the AI model best suited to the task to generate a response.
- The response is checked again before being returned to you, and, where the request involves an action rather than only a reply, executed only within the authorization you've granted.
3. Which Models We Use
NSOffice.AI supports multiple leading AI model providers, alongside a self-hosted model for certain workloads, so requests can be matched to the model best suited to the task rather than forced through a single provider. Every model — third-party or self-hosted — is independently risk-assessed and approved before it's used in production. See our Responsible AI Statement for how that governance process works.
4. How We Monitor Model Performance
"Model performance" isn't a one-time check at launch. Once a model is in production, we:
- Monitor its ongoing behavior for accuracy, safety, and drift from expected performance.
- Apply defined thresholds that trigger a re-evaluation if a model's behavior degrades or drifts from what was approved.
- Periodically re-assess approved models, and formally retire or replace a model rather than leaving it in production indefinitely without review.
5. Memory and Personalization
If you use the platform's memory features, the Service may retain certain facts such as your role, preferences, or context from prior conversations to make future interactions more relevant, rather than requiring you to repeat context every time.
- Memory data is stored within your organization's own regional environment and is never shared across organizations.
- Memory data follows the same deletion timeline as the rest of your data: a 30-day recoverable period, followed by permanent deletion within 180 days of removal or account closure.
- Memory reflects what you or your organization have used the platform for — it is not built from, or shared with, any other customer's data.
6. Output Limitations
AI-generated Output is produced probabilistically and may be inaccurate, incomplete, or unsuitable for your intended purpose. You're responsible for reviewing Output before relying on it for a business, legal, financial, or other consequential decision — the Service is a productivity tool, not a substitute for professional judgment. See our Usage Policy for more on acceptable use of AI-generated content.
7. Questions About a Specific Feature
If you want to understand how a particular feature handles your data — memory, a specific connector, or a specialized Bee — contact customersupport@nsoffice.ai, and we'll walk you through it directly.