The four requirements for enterprise AI
Twelve questions for any vendor.
Our answers beside each.
An open scorecard procurement can hand to any vendor, including us. A requirement is met when all three answers are yes.
01 · Control
Who can do what, with what data, and can you prove it?
| The question | Why it matters | Nexus's answer |
|---|---|---|
| Does one policy cover every prompt, agent and data source, in every tool? | Sixty tools mean sixty governance models. Govern the object, not the app. | Yes. Every chat, dashboard, skill, agent, knowledge source and tool connection carries a scope: Personal, Team, Project or Workspace. Roles follow the same logic. |
| Can you audit who accessed what, and who changed access, for at least a year? | The board will ask. So will the regulator. | Yes. Access Audit records who changed resource access, what changed and when, retained 365 days with a purge control. Stopped Turns and per-message token accounting complete the trail. |
| Is what an AI remembers about your people reviewed before it is kept? | Memory is where personal data accumulates quietly. | Yes. Memories carry a confidence score and wait as suggestions until approved, by default. Scoped recall, decay and expiry are settings. |
02 · Access
Will your people use the sanctioned tool, together, across functions?
| The question | Why it matters | Nexus's answer |
|---|---|---|
| Is the sanctioned tool the one your people would pick anyway? | Bans do not remove the choice; they hide it from the people meant to manage it. | Yes, by design. Chat with company context, @ / + mentions, AI Mode, quick actions, dashboards from a chat, the Nexus app on phones. Usage this week shows whether it is true. |
| Can a skill or agent built in one team be reused by the next without rebuilding it? | In personal tools and point solutions nothing is shared. The same connection gets built forty times, and the work breaks at every handoff between departments. | Yes. Publish a skill, share an agent, reuse a connection, by scope. The second person gets it free. Reuse rate is a number on Workspace Health. |
| Are connections approved once and used by each person with their own credentials? | The alternative is giving everyone the keys, or giving nobody anything. | Yes. "Shared workspace connection. Connect your account." Per-user OAuth; one person's token is never reused for another. |
03 · Safety
Where does your data live, who can reach it, and what leaves?
| The question | Why it matters | Nexus's answer |
|---|---|---|
| Does the workplace run in your own cloud account, with your data staying in it? | "Private cloud" means different things to different buyers. "In your account" is verifiable. | Yes. Deployed in your own cloud account. Knowbl's software runs there; your data does not leave it. |
| Are prompts and documents kept out of public model training and out of providers' own use? | The most common upload to external AI tools is source code. | Yes, under the standard Nexus architecture. The full data promise, with its qualifier, is on Security and Trust. We do not extend it. |
| Do you set retention, and are credentials encrypted and never shared between people? | Retention should be a setting, not a request to a vendor. | Yes. Retention, compaction and purge windows are settings. Credentials encrypted at rest and never returned by the API. Guest experiences default to deny-all origins and no tools. |
04 · Independence
If the vendor reprices or you part ways, what does it cost?
| The question | Why it matters | Nexus's answer |
|---|---|---|
| Is switching the model behind your workflows a setting, not a migration? | Model leadership and model pricing change quarterly. | Yes. Every stage resolves to a tier (Fast, Balanced, Powerful). Remap the tier in Stage Assignment and every stage follows. The registry accepts Bedrock, Vertex, direct APIs, OpenAI-compatible endpoints and custom HTTP. |
| Would your skills, agents, memory and connections survive a vendor change intact? | The context layer is the most expensive thing to rebuild, and the thing a fused stack holds hostage. | Yes. Skills, Knowledge, Memories and Tool Connections live in your workspace, in your account, independent of any model. |
| Are your data, prompts and outputs yours in the contract, with return terms at termination? | The architecture should match the agreement, and the agreement should match the architecture. | Yes. Your data is in your account and stays there. Termination and data-return terms are in the agreement, reviewed by counsel against every claim on this page. |
Stated plainly, so you do not have to ask: we are a small team by design, we do not train frontier models, and we rent the model on purpose. That is the architecture, not a disclaimer.