On this page
Short answer: "privacy-first" is not a standard. Ask five questions: where the data lives, who can read it, what leaves your network, what the AI features send, and how access is controlled and audited. A vendor that answers plainly is usually the safer choice.
What it should mean
Project data is more sensitive than it looks. Plans name clients and prices, risk registers describe what could go wrong, and issue trackers describe unreleased work. Privacy-first should mean you decide where that data lives and who can reach it, and the vendor does not quietly collect more than it needs.
The five questions
1. Where does my data live?
Can you choose the location, including your own servers? For regulated work, residency matters as much as encryption. A tool you can host yourself answers this best. A vendor-hosted tool should name the regions it uses.
2. Who can read it?
Look at roles and scope. Can you limit people to the projects they need? Is there a read-only role? Can outside people be given limited access without a full account? Are administrative actions logged?
3. What leaves my network?
Ask about telemetry, licence checks, error reporting and third-party scripts. Get the list of exactly what is sent. A good answer is short and specific.
4. What do the AI features do with my data?
This is now the most important question. Ask which providers are used, whether your data is used for training, whether you can use a local model, whether you can turn AI off entirely, and whether the estimated cost and content are shown before anything is sent.
5. How is access controlled and audited?
Look for single sign-on, two-step verification, password policy, session control and an audit log you can export. An audit trail that cannot leave the product is not much use to a security team.
How NUSK answers them
- Location. Self-host with Docker or a desktop installer on your infrastructure. Data stays on the server or PC you choose. A hosted edition also exists.
- Access. Five roles (admin, PM, team lead, member, viewer) enforced on every endpoint. Per-project permissions and custom roles. Read-only guest links for outside people. Deactivating a user takes effect on their next request.
- What leaves the network. A licensed on-premise server checks in with the vendor and sends only an installation ID, a machine fingerprint, the host name, the version and the active user count. It never sends project data. Without a valid licence the workspace becomes read-only; viewing, export and backup keep working.
- AI. Choose OpenAI, Anthropic, Azure OpenAI, Google Gemini, or any OpenAI-compatible or local server such as Ollama, vLLM or LM Studio. Keys are stored encrypted and never shown again. With AI off, nothing is sent anywhere. Larger analyses show their estimated cost first, and AI changes are proposals you confirm, reject or undo. Limits and usage reporting are available per instance, project, team, user or role.
- Identity and audit. Single sign-on (Google, Microsoft, or any SAML 2.0 provider such as Okta, Entra ID, ADFS or OneLogin), SCIM provisioning, authenticator-app two-step verification with recovery codes, a password policy, login lockout, a device list with remote sign-out, and an audit log with filters, CSV export and streaming to a SIEM. Every change to a work item keeps a field-by-field history.
A one-page questionnaire you can copy
- Where is my data stored, and can I choose?
- Is data encrypted in transit and at rest? How are keys managed?
- What exactly does the product send to the vendor or third parties?
- Which AI providers are used? Is my data used for training? Can I turn AI off or use a local model?
- What roles exist? Can I scope access per project? Can I give read-only access to outsiders?
- Is there SSO, MFA and a password policy?
- Can I export the audit log and stream it to my SIEM?
- How do backup and deletion work? What happens when I leave?
- Which certifications or audits does the vendor hold? Ask for the reports.
- How are vulnerabilities reported and fixed?
When a cloud tool is fine
A well-run cloud tool with the right controls is acceptable for many teams, especially without residency rules. The point of the questions is to find out, not to rule it out.
What we do not claim
We do not claim a security certification we have not earned. If a customer needs a certified report, ask us what exists today and treat it as a requirement to verify.
Try NUSK with AI switched off · Run it on your own servers
See it with your own project
Import a plan or a Jira project into a free NUSK workspace and judge it on your own data.