Data, security and AI
What ManufactINS does with your plant data, and how its AI is used — as the platform actually works.
Data and security
Each person sees only their own plants and orders
Every account has one of ten manufacturing roles, or Observer, which only reads, and a list of the plants and orders it may see. Pages and requests check both. Observer, presenter and walkthrough accounts can read but never change anything. When the owner removes an account or changes its access, its session ends at once.
Sign-in
Passwords for accounts created in the platform are kept only as scrypt hashes, never as text. The owner can sign in with a passkey. Sign-in attempts are limited per address and per account, and repeated failures lock the account for a while. A company account can be given an end date. A session is a signed cookie the browser’s scripts cannot read, and it lasts at most 12 hours.
Keys and passwords kept encrypted
The AI service keys, the mailbox password and the credentials for connected systems are encrypted on disk (AES-256-GCM) with a key held by the server.
A record of who did what
What administrators do is written to a log that is only ever added to. The platform’s audit trail cannot be edited or deleted, and its chain can be verified. A confirmed decision keeps the record before and after, and the order-to-cash ledger keeps every correction with who made it and why. Page views and actions of signed-in accounts are recorded; what you type is not. Exporting records and reports needs the export permission.
Where your data lives
On the server your ManufactINS runs on, in its own data folder. The owner can take a backup from Admin, with secrets left out; when cloud backup is set up, a copy goes there too. A plant or order can be archived, and deleted only after its code is typed to confirm.
Cookies
Only the ones the platform needs to work: your sign-in, your language, the time zone you chose for showing times, and the company sign-in steps. No advertising or tracking cookies.
How the AI is used
The AI helps; people decide
It reads documents, drafts replies and reports, checks figures and recommends. Approvals, quality releases, payments and safety decisions stay with the person responsible.
ManufactINS’s own AI models
ManufactINS runs on its own AI models, fine-tuned on manufacturing work, with the strongest model for each task: reading documents, handwriting from the floor, the hard questions and the quick replies. Leading language models are used only where they improve a result, under the same checks.
What is sent to it
Your question and the parts of the plant records your account may see that bear on it — not the whole database. A file you add to a chat is read for that chat.
Where it goes
To ManufactINS’s own models, on the servers your deployment runs them on. A task that uses an outside language model may be processed outside Egypt, under that service’s terms. A deployment can keep every task on ManufactINS’s own models.
Voice
Voice calls are set not to save the audio or keep the transcript, and ManufactINS stores no audio from calls, read-aloud or dictation. Dictation can also use your browser’s own speech recognition.
What runs without a model
Ask ManufactINS answers from the plant records and its own knowledge base, and typed or spoken floor updates are read by the platform’s own rules in English and Egyptian Arabic. Dashboards, registers, the order-to-cash ledger, reports, exports and drafts from templates run on ManufactINS’s own engines. Reading photos and scans, model-written answers and agent goals use the AI models.
What it does not do on its own
A data input, a floor update, a document it read or an action it plans waits until a person confirms it. Role agents prepare each decision for the person responsible; only when the owner sets an agent to act does it carry out internal, reversible steps and confirm what a filed document already answers. Everything else waits for the person responsible. The owner can let an agent work on a goal within the permissions of the person it acts for. It sends nothing outside the platform: a message or report goes out only when a person approves or sends it. A read-only account’s AI can only read.
Its limits, and how they are checked
It can misread a document, miss context or use old figures. So every answer names its sources and says when the records do not hold the answer. Check important figures before you act. It is not engineering, legal or financial advice.
What not to upload
Passwords, personal data you may not share, and documents you have no right to share. Check your contracts’ confidentiality terms before adding a customer’s files.
Reporting a security issue
Tell the administrator of your ManufactINS, or write to manufactins@epsilonaiglobal.com. Please do not make it public before we have had time to fix it.