A CTO asked us last month: "If the AI learns from all my sessions, how do you stop Client A's information from leaking into Client B's experience?" It is the right question. Here is how the architecture actually works.
Coaches who evaluate CoachNova tend to get stuck on the same tension. On one hand, they want a digital twin that sounds like them, that captures their coaching style and gets sharper over time. On the other hand, they need absolute confidence that what one client shares never reaches another client's context. These two goals seem like they should conflict. They do not. But only if the architecture is built to separate them from the start.
Four layers, one principle
Everything CoachNova's AI does runs through four layers. Each layer has different access rules, different data, and a different purpose. The principle underneath all of them is simple: your coaching voice is yours. Your clients' information belongs to each client individually.
Layer 1: CoachNova Rules
This is the foundation. It contains the ICF core competencies, ethical guidelines, and platform-wide configuration that apply to every interaction across every coach and every client. Think of it as the guardrails. No matter what the AI generates, it must align with professional coaching standards. This layer never contains any coach-specific or client-specific information.
Layer 2: Coach Signature
This is your digital twin voice. It is built from your style questionnaire answers, the coaching frameworks and documents you upload, and the edits you make to AI-generated outputs over time. When you correct a session summary or adjust a reflection prompt, the system learns more about how you think and communicate.
This layer is shared across all your clients, because it represents you, not them. It contains patterns like "this coach tends to ask open questions about values before exploring behaviour" or "this coach uses metaphor frequently and avoids directive language." It does not contain anything about Sarah's leadership challenge or James's career transition. It captures how you coach, not what your clients tell you.
Layer 3: Client Context
This is where strict separation lives. Each client has their own isolated context containing their session history, goals, commitments, progress notes, and messages. When the AI prepares for a session with Sarah, it has access to Sarah's context and nothing else. James's sessions, goals, and conversations do not exist in that preparation. They are not filtered out. They are simply never included.
This is an important distinction. Some systems filter sensitive information after combining data. CoachNova never combines it in the first place. The contexts are separate by design, not by post-processing.
Layer 4: Current Context
This is the specific input being processed right now. A transcript from today's session, a question you are asking the AI, a reflection prompt being generated. It is temporary and scoped to the current task.
How the digital twin improves without crossing boundaries
This is the part that matters most, and the part that most people find counterintuitive at first.
Your digital twin voice (Layer 2) gets better with every session you run. After twenty sessions with ten different clients, it has a much richer understanding of your coaching patterns than it did after your first session. That is the whole point. The more you use CoachNova, the more it sounds like you.
But here is what it learns: style, not substance. It learns that you tend to reflect back emotions before asking about actions. It learns that you use specific frameworks for leadership transitions. It learns your vocabulary, your pacing, your preference for certain types of questions. None of that is client-specific. It is coach-specific.
What it does not learn across clients: that Sarah is struggling with her board. That James is considering leaving his role. That your Tuesday client disclosed a conflict with their direct report. All of that stays locked in each client's own context (Layer 3) and is never used to inform another client's experience.
The result: your twentieth client gets a better version of your coaching voice than your first client did. But they get zero information about any other client.
Your coaching voice improves with every session. Your clients' data never crosses over.
"Does CoachNova train on my clients' data?"
No. This is worth addressing directly because it is the most common concern coaches raise, and the answer is unambiguous.
CoachNova does not use your session data to train foundation AI models. Your transcripts, session notes, and client information are not fed into any general-purpose machine learning pipeline. They are not used to improve the product for other coaches. They are not shared with third-party AI providers for training purposes.
Your data is used for one thing: making your CoachNova experience better for you and your clients. That means building your digital twin voice (Layer 2) and maintaining each client's context (Layer 3). That is it.
This is not a policy choice that could change with the next terms of service update. It is an architectural decision. The system is built so that coach and client data stays within the coach's own environment.
What this looks like in practice
Here is a concrete example. You have three clients: Sarah, James, and Maria.
On Monday, you finish a session with Sarah about managing upward to a new CEO. CoachNova generates session notes and a reflection prompt for Sarah. The AI draws on your coaching voice (Layer 2) and Sarah's session history (Layer 3). James and Maria do not exist in this context.
On Wednesday, you prepare for a session with James about a team restructure. The AI reviews James's previous sessions, goals, and commitments. It uses your coaching voice to draft preparation notes in your style. Sarah's CEO challenge and Maria's information are not part of this preparation. Not filtered out. Not present.
On Friday, you review a between-session check-in that Maria completed through her digital twin access. The AI generates a summary using your coaching voice and Maria's context. Sarah and James do not factor in.
Your voice is consistent across all three. The information is completely separate.
For the technical stakeholders
If you are the CTO, IT lead, or compliance officer evaluating CoachNova on behalf of a coaching practice or organisation, here is the summary:
- Data isolation is architectural, not policy-based. Client contexts are separate data structures, not filtered views of a shared dataset.
- The coach signature layer contains no client information. It stores coaching patterns and style characteristics, not details from individual client sessions.
- No cross-client data access is possible in the context creation pipeline. The system assembles context per-client from isolated sources.
- Session data is not used for model training. No data leaves the coach's environment for ML pipeline purposes.
- All infrastructure is EU-hosted. For details on hosting, sub-processors, and data processing agreements, see our Trust Centre.
If you have questions that go beyond what is covered here, we are happy to walk through the architecture in more detail. Reach out through our contact page or ask to schedule a technical review call.
Try it free with your first client.
Full access, no credit card. EU-hosted. Your data stays yours.