Security you can verify, not just trust.

The centre of a Performance Hub deployment is an appliance at your facility - an Edge Processor for video, or an AI Agent device. Performance Hub is edge-first, not edge-only: we process sensitive workloads on that appliance where practical, while cloud services provide account management, user and facility data, configuration, operational metadata, logs, selected storage, and experiences that are delivered through the cloud platform.

• The AI Agent's conversations, memory, workspace files, wiki, skills and schedules are stored on the device you own and physically control. They are not held in a shared cloud service.
• Video is processed at your facility; the cloud stores, serves and manages. Recording happens on the appliance first, so a cloud interruption never causes a gap in recording.
• The appliance holds only a transient local buffer for footage; its home is the cloud region you choose.

The result is a deliberate split. Workloads such as agent state, live video and on-device analysis can stay close to where the data is created, while cloud-held data remains scoped to your account and governed by the platform's access controls, retention settings and configured storage regions where applicable.

This is the headline posture, and it is a genuine architectural property of the appliances.

• No inbound ports, no firewall holes. The appliance opens the connection to the cloud itself over encrypted transport. There is no port-forwarding, no dynamic DNS, no VPN appliance and no inbound firewall exceptions to configure. The attack surface that plagues traditional DVR/NVR installs is simply not there.
• Commands reach the device over its own connection. The platform can send commands to a device - for example, an authorised user rebooting an appliance at another facility they manage - over an authenticated outbound management channel the device established. Nothing is opened on your network to make that possible; the device receives instructions on a channel it opened, authenticates them, and acts.
• Cameras are LAN-isolated. Cameras talk only to the Edge Processor on the local network. They never need internet access, never expose ports to the world, and their credentials live only in the appliance's configuration. A compromised camera brand's cloud is irrelevant: your cameras are not on anyone's cloud.
• Remote access via tokenised tunnels. There is no directly reachable public endpoint on the device. Remote access flows through authenticated, tokenised tunnels with per-device credentials.
• LAN access is protected too. The direct-access path on your local network is TLS-protected and requires a valid Performance Hub session token.

Backup encryption keys are generated at pairing and held by Performance Hub on your facility's behalf. They are never stored inside the backups themselves and never exposed in the browser, so the storage provider holds ciphertext only.

• The device holds no long-lived secrets to your data. It authenticates with short-lived tokens derived from a per-device secret issued at pairing.
• There is no standing API key for AI models on the device.
• A leaked device identifier alone is useless, because there is no secret to sign with; a leaked token expires quickly.
• Removing a device revokes its credentials immediately. A removed device can no longer reach your Performance Hub data, models or notification channels, whatever state it is in. The device is factory reset and its cloud backups are retained under your facility and retention policy.

• Authentication. Users authenticate via federated single sign-on (Google, Microsoft, Apple, Slack) or email-based login. Organisation policy can require additional Performance Hub multi-factor authentication, independent of MFA enforced by the identity provider. There are no standing passwords to the device; sessions use short-lived signed tokens minted when you open the module.
• Authorisation is role-based, scoped to facilities, regions or the whole organisation, and enforced centrally on every request.
• Attribution. Interactive chat messages and actions carry the signed-in user's identity. Actions an agent runs on its own are attributed to that agent, so activity is always traceable to a person or the agent that performed it.
• Grant attenuation. No one can grant the agent, or another user, more scope than they themselves hold.
• Support access. At your request, for support or product-architecture work, access for Performance Hub staff and organisation administrators may be granted through organisation settings and/or individual agent configuration. Architecture advice may also include on-site or remote sessions.

Every customer record is scoped to a facility, which rolls up to a brand and an organisation. Authorisation is enforced centrally on every API call, resolving the facility-to-brand-to-organisation hierarchy and verifying the caller's role covers it.

• Cameras and footage never blend across facilities; a user assigned to three of an organisation's ten facilities sees exactly those three.
• Facial collections are isolated per facility by default. Organisations may enable cross-facility access to facial collections for multi-site deployments under their policy.
• The AI Agent device is scoped by default to the facility it is paired to.

Beyond the product architecture, Performance Hub operates continuous security practices on the platform itself:

• Audit logging. Account activity in our cloud environment is recorded with an organisation-wide, multi-region audit trail, with log-file validation and encrypted log delivery.
• Centralised security posture monitoring. We continuously aggregate and prioritise cloud security findings and track posture across the environment.
• Perimeter and vulnerability scanning of our public-facing services, on an ongoing basis.
• Software composition and dependency monitoring, so the components the platform is built from are tracked and kept current.
• Managed, staged updates. Appliances update from the cloud in a controlled window (a nightly maintenance window for video appliances; a user-gated "update now" for AI Agent devices). Updates replace the software, not the data, and state survives updates.
• Backups and disaster recovery. Encrypted, hourly, incremental backups to customer-chosen regions, with full-restore safety backups and device-replacement seeding.

The following boundaries are part of the security model and are stated so they can be assessed accurately:

• Default SaaS tenancy uses logical isolation via centralised access control. Dedicated architecture is available for eligible enterprise and government customers by contract (see Multi-tenancy above).
• Local data on the appliance remains under the physical control of your premises. Encrypted backups leave the device before upload; removing a device revokes its credentials immediately.
• Secrets, credentials and other confidential material placed on an AI Agent by your organisation (including through scripts, skills or workspace files) are under your control. Performance Hub encrypts and securely stores agent backups; exposure introduced by customer-configured agent behaviour is your responsibility. See Shared responsibility.