Cerebro holds what your people know. NovaCube holds your records. Here is how both are kept, who can see them, and what we will never do with them.
Each customer gets its own database. There is no shared table that a missed filter could leak across companies. A test suite that tries to read one company's data as another runs before every release.
Every fact is marked private, team, department or company, and can carry a clearance level. The rule is applied in the database query itself — a manager's screen physically cannot fetch a colleague's private note.
Each search, read and change is written to a log that can be added to but never edited. It records who, when and which items — not the text of the question. An admin who opens a private note must give a reason, and that reason is logged too.
Cerebro and NovaCube run on servers in Kuala Lumpur (Shinjiru Technology), on separate machines. Data is encrypted on the way to and from your phone or laptop.
A copy of every company database is taken nightly, encrypted before it leaves the server, and stored on a second Malaysian server. We have restored from these backups end to end — not just assumed they work.
To answer a question, the AI model receives the question and the handful of memories needed to answer it — not your whole database. Searching, matching and voice-to-text run on our own servers. NovaDyne never trains any model on your data, and our contract says so.
People sign in with a link sent to their work email, or with their Microsoft 365 or Google account. Cerebro stores no passwords. Each phone and laptop is a separate connection you can switch off.
One click switches off their phone, laptop and AI connections together. What they shared with the company stays; their private notes can be reviewed by an admin, with a logged reason, and consolidated or deleted.
Every company has an AI allowance with a hard ceiling. When it is used up the AI features pause until you top up — searching and saving keep working. There are no surprise bills.
Your workspace is one database, so your export is a complete copy of everything — memories, documents, decisions, history. On exit you get the full export and certified deletion within 14 days.
These are written into every contract, not just this page.
Copy this into your vendor checklist. If a row is missing, ask — we would rather answer than let you guess.
| Tenancy | One PostgreSQL database per customer. No cross-database queries in the application; only the migration runner and the billing meter touch more than one. |
| Access model | Private / team / department / company scopes plus a 0–3 clearance level per item, enforced in SQL. Directors and admins can read shared scopes; nobody can browse another person's private items without a logged reason. |
| Audit trail | Append-only log per company: actor, timestamp, action, item ids, hashed query. Available to your admin in the app. |
| Hosting | Shinjiru Technology Sdn Bhd, Kuala Lumpur, Malaysia. Cerebro and NovaCube on separate servers; Cerebro on its own hardware, store and runtime. |
| Encryption | TLS 1.2+ with HSTS in transit. Bearer tokens stored as SHA-256 hashes; third-party integration tokens encrypted at rest (Fernet). Backups GPG-encrypted before leaving the server. |
| Backups | Nightly, encrypted, copied off-box to a second Malaysian server; restore procedure tested end to end. |
| Authentication | Magic link by email, Microsoft 365 (Entra ID) or Google sign-in; OAuth 2.1 with PKCE for AI connectors; per-device tokens with expiry and one-click revocation. No passwords stored. |
| AI processing | Answers: an AI model provider's business API (currently DeepSeek; Anthropic Claude on request) receives the question plus the retrieved memories. Embeddings (bge-small) and voice-to-text (Whisper) run on our servers. Voice replies, if a person sends a voice note, use Microsoft's speech service and can be switched off. Dedicated or on-premises deployment with a model of your choice is available. |
| Rate limits & caps | Per-IP and per-token rate limits on every door; per-company AI allowance with a hard ceiling; per-company document quota. |
| Data residency & law | Data stored in Malaysia. NovaDyne Sdn Bhd acts as data processor under the Personal Data Protection Act 2010; governing law Malaysia; a data processing agreement is available. |
| Export & deletion | Full database export on request; certified deletion within 14 days of contract end. |
| Incident notification | Affected customers informed within 72 hours of confirming a breach. |
| Certifications | None yet — Cerebro launched in 2026. We publish this page instead of a badge, and will update it when an audit is complete. |
Every NovaCube + Cerebro pilot runs 60 days against four numbers agreed at kick-off. At day 45 both sides sign one scorecard. Case studies will appear here as pilots complete, with each customer's permission.
Quotations, invoices, reports and messages produced with the AI's help during the pilot — counted, not estimated. Default target: 20.
The share of pilot users active every week, from the system's own logs. Default target: 70%.
Hours per person per week saved, assessed by your sponsor, not by us. Default target: 6.
Facts, decisions and procedures stored in Cerebro, and recalls your people rated useful. Default targets: 100 captured, 10 useful recalls.
Three of four is a pass. The pilot fee is credited in full against the first year if you continue. The first pilots started in September 2026; when they complete we will publish their scorecards here, numbers first.