Trust
Security and deployment
AI team members act inside business systems, so the controls around them have to be at least as strong as the ones around a person doing the same job.
Access and accountability
Single sign-on
Microsoft Entra ID and Google sign-in. Only the providers an organisation has configured are offered on its login page.
Multi-factor authentication
Time-based one-time codes, enforceable per organisation.
Roles and rights
A rights catalogue, custom roles, and per-member overrides. Team members hold their own scoped grants to connectors.
Audit trail
Jobs, approvals, instruction changes and administrative actions are recorded with who did them and when.
Your data
Tenant isolation
Each organisation is a separate tenant. Row-level security in the database enforces the boundary, not only the application.
Secrets encrypted at rest
Connector credentials and API keys are encrypted before they are stored. They are decrypted only when a team member needs them for a call.
Your model, your terms
Choose the model provider per organisation. In the cloud that is Azure OpenAI by default; on your own servers it can be a fully private model.
Two ways to run it
OvermindAI cloud
- Hosted on Microsoft Azure.
- One tenant per organisation, isolated at the database level.
- Azure OpenAI models by default. Prompts and documents are not used to train models.
- Managed backups and updates.
On your own servers
- Runs as containers on a server you control, behind your own single sign-on.
- Indexes network file shares read-only, so nothing leaves the building.
- Uses a model server on your network, or a cloud provider if your policy allows.
- Licensed per organisation.
Questions a security review needs answered? Ask us directly and we will go through them with your team.