Security questions deserve direct answers.
Review where source code lives, what an AI provider can receive, how credentials are handled and when Nova needs a person to approve an action.
The questions your review will ask.
- Where is the working copy?
- In the engineering environment selected for the project, either customer-owned or managed by NovaCore.
- Does any code reach an AI provider?
- Yes. Relevant source excerpts and tool results may be included in model input. The full repository does not need to move.
- Does the AI provider receive credentials?
- No. Credentials are not included in prompts or model responses.
- How is connected access limited?
- Repository and service access is scoped to the projects and actions Nova needs for the assigned work.
- What decides whether an action happens?
- Your configured rules determine whether Nova may proceed, must ask a named approver or is blocked.
- Can an approval be reused?
- No. Approval applies to the specific action and target shown to the reviewer.
- How are customers separated?
- Customer data is logically isolated, and sensitive stored content is encrypted.
- Can we see what happened?
- Nova retains a visible history of the work, approvals, usage, verification and final result.
What Nova enforces.
Every request is scoped to one customer, and stored data remains separated between customers.
Sensitive content is encrypted in storage and protected in transit.
Nova sends the configured AI provider only the context needed for the current task; credentials stay out.
Sensitive actions pause for the right person, while prohibited actions remain blocked.
Nova records what was requested, what was approved and what the connected service returned.
Customer-owned and managed environments have explicit access, network and operational responsibilities.
What stays put, and what crosses.
Repository checkout, build caches, project toolchain, browser state and narrowly scoped development access.
Task instructions, relevant excerpts and bounded tool results when needed for the work.
Work history, approvals, usage, project guidance and selected verification evidence.
Only the scoped credentials and actions configured for Jira, GitHub or another approved service.
“Your code never leaves” would be misleading for a product that can use hosted AI providers. The repository stays in the selected engineering environment, while relevant excerpts and results may be sent for the current task. Credentials are not sent to the AI provider.
What we are not claiming.
AI can still be wrong
Rules limit authority; they do not make reasoning infallible. Verification and human review remain part of the operating model.
Allowed context is still disclosure
If a source excerpt is sent for AI processing, the configured provider receives it. Provider terms and retention therefore matter.
An approved action can have impact
Approval means the action is allowed, not risk-free. Scope, service permissions and downstream controls still matter.
The work environment is powerful
Isolation, patching and network reach depend on who operates it. Managed and customer-owned environments have different shared responsibilities.
Shared service or dedicated deployment.
A shared NovaCore service with customer isolation and encrypted sensitive content. Work can run in your engineering environment or one managed by NovaCore.
Dedicated deployment, private networking, tailored retention and security review are available where the operating requirement justifies them.
Ask about a required certification or control framework early. We will describe current evidence and gaps directly rather than using a generic trust badge.
Send this to your security team.
Then bring back the questions about AI-provider data flow, environment responsibility, credential custody and approval rules. Those are the useful ones.