Skip to content
Security and Trust

Everyone in.
Nothing out.

Every employee working with AI. No data leaving your account. Written for the CISO and the General Counsel, with the controls named as they appear in the product.

The data promise

Your data stays yours.

Nexus is designed so your enterprise data remains private, protected, and under your control. When using the standard Nexus architecture, your prompts, documents, and model interactions are securely processed without being used to train public AI models or shared with underlying model providers for their own use. You get access to powerful AI capabilities while maintaining enterprise-grade control over your data.

Your data powers your AI experience, not someone else's model.

The same words, with the same qualifier, appear in the CISO packet, in our proposals and in the General Counsel FAQ. Compare them.

With Nexus
  • Your data remains private and protected
  • Your prompts and outputs are not used to train public AI models
  • Your enterprise data is not shared with model providers for their own purposes
  • Data is encrypted in transit and at rest
  • Access is governed through enterprise security controls
  • Nexus provides a secure orchestration layer between your data, applications, tools, and AI models
The controls

Where data lives. What can touch it.

Hosting

In your own cloud account

Knowbl's software runs in your account; your data does not leave it. We claim "in your account", never a residency or compliance your deployment has not earned.

Identity and access

Per-user OAuth, scopes, roles

Connections approved once by an admin, authorized by each user with their own account. One person's token is never reused for another.

Encryption

In transit and at rest

Data encrypted in transit and at rest. Model registry credentials encrypted at rest and never returned by the API.

Tool governance

Allow-lists and declared tools

MCP allow-lists per assistant and per guest experience. Each subagent declares the exact tools it may call. Timeouts and a circuit breaker on every server.

Memory

Asks before it remembers

Memories wait for review before they are kept, by default. Recall is scoped. Decay and expiry on schedules you set. Admins curate.

Retention

Windows you set

Retention, compaction and purge windows are settings, not requests to a vendor. Access Audit records kept 365 days with a purge control.

Audit

Who changed what, and when

Access Audit by resource and member, with revoked grants and activity events. Stopped Turns on conversations. Token usage on every message.

Guest experiences

Least privilege in front of the public

Deny-all embed origins and no tools by default. Signed visitor identity, tag-scoped knowledge, human handoff to a Live tab.

Compliance

SOC 2 Type 2

Cited within its certified scope. Our controls support your compliance program; you own it. Counsel reviews every data and IP claim before publication.

The CISO packet

Forward it before the meeting.

Hosting architecture in your account. The OAuth model. Encryption. Retention. MCP allow-lists. The audit trail. SOC 2 Type 2. The General Counsel FAQ travels with it.

Sent by a practitioner, not a sequence. No follow-up unless you ask for one.

Questions we get

What CISOs and General Counsel ask.

Where does our data live?
In your own cloud account. Knowbl's software runs there; your data does not leave it.
How do connections to our systems work?
Each user authorizes with their own credentials. Credentials are encrypted at rest. Connections run through allow-lists, and each agent declares the tools it may use.
What can we audit?
Access and activity, retained for 365 days, at the level of the object. Who changed access, what changed, when. Stopped Turns on conversations. Token usage on every message.
Does Knowbl or any model vendor train on our data?
Not under the standard Nexus architecture. Your prompts, documents and model interactions are processed without being used to train public AI models or shared with underlying model providers for their own use. Data is encrypted in transit and at rest, and access is governed through your enterprise security controls.
What stays ours in Nexus?
Your data, your prompts and your outputs are yours under the agreement, and they sit in your own account. Final wording follows counsel's review against the customer agreement.
What happens if we terminate?
Your data is in your account and stays there. Termination and data-return terms are in the agreement, and the account team walks through the options with you.
What stops an agent from doing something it should not?
It cannot see what it was not scoped into. It cannot use a connection the user did not authorize. A subagent declares the exact tools it may call and cannot exceed them, with an iteration cap. Failing tool servers trip a circuit breaker. Scheduled runs require configuration before they run. Humans pick up handoffs in Tasks.
Our people will paste customer data into it. What does it remember?
Memory is extracted with a confidence score and, by default, held for review before it is kept. Recall is scoped to the user and project unless you choose workspace-wide. Facts, preferences and procedures decay and can expire on schedules you set. Admins can browse, approve, dismiss and delete.