DEEPSELF & MANTRA
DEEPSELF & MANTRA
Making a DeepSelf means giving us access to the most sensitive thing your organisation has: the knowledge in its people's heads, most of which has never been written down. It is held on your own AI platform throughout, we reach it through credentials you can revoke at any time, and when the work ends none of it remains with us.
The choice between them can follow your own security and compliance requirements, because the method works the same on each.
Your data stays inside your Azure tenant and is never sent back to OpenAI for training. The work sits under the Microsoft enterprise agreement and compliance terms you have already approved.
Claude and other models run inside your own AWS account, so your data never leaves it and the security controls and compliance certifications you already hold cover the work.
Models run inside your own Google Cloud project, under the data governance controls and security posture you already have in place.
A dedicated instance with a contractual guarantee that your data is not used to train models, backed by Anthropic's enterprise security and compliance documentation.
Open-weight models running entirely on your own premises or in your private cloud, for the most sensitive environments. An air-gapped setup, with no connection to the outside internet at all, is possible.
The platform is yours, and the connection to it is ours to secure. The machines and the credentials we use to reach it are set up the same way for every client.
All of your work happens on a hardened machine kept for it alone, carrying no personal applications and no other client's material. The computer that touches your documents does one job only.
Its disk is fully encrypted and its firmware protected. It takes security updates automatically, locks its screen after a short idle, and can be wiped remotely if it is ever lost or stolen.
We never reach your systems from a public network. The connection is always a secured, trusted one with the traffic encrypted in transit, so a hotel lobby or a café is never part of the path between us and your data.
Logging in to anything critical requires a hardware security key: a small physical device that proves who is logging in by cryptography, cannot be phished, and cannot be copied from a distance. If the key is not physically present, the login does not happen.
Your credentials are kept in a vault used for your work and nothing else, with a different generated password for every point of access. Passwords are retrieved from the encrypted vault rather than typed, so they never pass through a keyboard or a clipboard.
Both ends of the work follow a written sequence, so your security team knows exactly what has been set up and, later, exactly what has been taken down. At the end you revoke our access yourself, the vault holding our credentials is deleted, and no copy of your material remains with us in working files, archives or backups.
01 Legal agreements signed
02 You provision dedicated access credentials
03 Workspace access set up inside your environment
04 A credential vault created for your work alone
05 Every point of access documented in writing
01 Written confirmation that the work has concluded
02 You revoke all workspace access
03 You deactivate the credentials you provisioned
04 The credential vault archived and scheduled for deletion
05 Written confirmation that our data obligations have been met
Each of these commitments is written into two agreements we sign before the work starts, so they bind us as terms of a contract.
→ A precise definition of what counts as confidential
→ Binding obligations on how it is handled and protected
→ Clear exclusions for information that is already public
→ A term of several years that outlasts the work itself
→ A clause requiring the return of your materials at the end
→ Processing documented to GDPR requirements
→ A clear statement of who is controller and who is processor
→ The security measures we must maintain, spelled out
→ A process for authorising any sub-processor before it is used
→ An obligation to delete your data when the work ends
The Standard
Everything the work makes stays where it was made.
What the work surfaces is the judgement that walks out of the door when a key person leaves, and the pattern recognition nobody has managed to write down. A competitor cannot get those by hiring your staff or reading your website, and putting them into language is what makes a DeepSelf possible. Once they exist in language they can be copied, which is why the work is only ever done on systems you control.