Where Your Data Lives, and Who Can Touch It

The baseline is encryption everywhere, at rest and in transit. Every store that holds your data, the files, the extracted text, the knowledge graph, the search index, and the application records, encrypts that data at rest using AES-256, which is the standard used to protect sensitive data across the industry. Every movement of data, from your browser to the application and between the platform's internal services, travels over encrypted transport using current TLS, so your documents are protected both while they sit and while they move. Unencrypted transport of client data is prohibited outright rather than treated as an acceptable shortcut. None of this is exotic, and that is the point: the protection of your data rests on well-established standards applied consistently, not on a clever scheme that has to be taken on faith.

The protection that matters most for a competitive business is isolation, the assurance that your data and another contractor's data never mix, and the platform is built so that isolation is architectural rather than a matter of careful filtering. Each client's knowledge graph lives in its own separate database instance. Each client's search index lives in its own separate namespace. Each client's files are stored under paths scoped to that client alone. There is no query path in the system that reads across clients, because the separation is built into where the data physically lives, not bolted on as a rule that code has to remember to follow. Tenant identity is resolved from an authoritative source and required on every read and write, and a request that cannot establish its tenant fails rather than guessing.

This design decision reflects how the platform treats your data by default. All client data is classified as confidential from the moment it arrives, and it is never downgraded without your written consent. We build the isolation into the architecture so that you can put your bid strategy and your pricing history into the platform without wondering whether a competitor on the same system could ever reach it, because the answer is not "our policy forbids it," it is "there is no path for it."

A specific and reasonable fear about any AI platform is that your documents will be absorbed into some model's training and quietly resurface in an answer given to someone else. On this the commitment is direct: your documents are never used to train AI models. The platform uses AI services under commercial terms that prohibit the provider from training on submitted data, and your content is never sent to consumer-grade or free-tier AI endpoints where such protections do not hold. Your documents are processed to answer your questions, and then they remain yours. This is the single most important promise for a firm whose documents encode its competitive edge, and it is a promise about architecture and contracts, not merely about intentions.

The outside services that do process data in order to make the platform work, for parsing, extraction, and answering, are disclosed to clients as a named list rather than hidden, and clients are notified of material changes to it. Transparency about who touches the data is part of earning the trust, because a protection you cannot inspect is a protection you have to take on faith, and the goal here is to take as little on faith as possible.

Protection also means limiting who inside the platform can reach anything, and what happens when you leave. Access to client data requires multi-factor authentication without exception, is granted at the minimum level a role actually needs, and runs through named individual accounts rather than shared logins, so every access is attributable. Credentials and keys are held in a dedicated secrets manager and injected at runtime rather than written into code or shared over chat, access is reviewed on a regular cadence to strip out anything stale or over-broad, and departures trigger prompt revocation. No third party is given standing access to your data, and any time-boxed support access is approved deliberately and revoked when the work is done.

When the relationship ends, your data does not linger. On wind-down, all of your tenant data, the documents, the graph instance, the search namespace, the stored files, and the application records, is deleted within thirty days, and that deletion is confirmed to you in writing on request. Deleting an individual document during normal use cascades to everything derived from it, the chunks, the graph nodes, the vectors, so that removing a document actually removes it rather than leaving fragments behind. We build the exit to be clean so that you can adopt the platform without feeling that your data has been captured by it.

Security is a practice, not a finished state, and honesty requires saying that plainly. The controls described here are real and they are applied consistently, and no responsible provider claims that any system is perfectly and permanently secure, because the threats change and the work of defending against them never ends. What can be promised is a specific and inspectable posture, built on encryption on well-established standards, isolation designed into the architecture, a firm commitment never to train on your data, a disclosed list of who processes it, disciplined access control, and a clean, confirmable deletion path. The subprocessor list is shown rather than hidden precisely because trust should rest on what you can verify, not on a badge. The commitment is to a serious, continuously reviewed practice, and to being straight with you about it.

Protection is what makes the platform safe to use. The reason to use it is everything that protection is guarding, which is the considerable value locked inside documents you already own and mostly cannot reach. What that value actually is, and how a well-built knowledge base compounds it over time, is the subject of the next article.

Axion - Institutional Intelligence for Solar, BESS & Construction

Follow us on Axion for weekly technical insight.