Trust & Data
ARKTOR can use a local model on suitable hardware, supported frontier AI, or a hybrid setup. The deployment choice changes where data goes. Permissions, execution and verification remain separate concerns.
Where data goes depends on the deployment.
- Local: a suitable local model can run on customer-controlled hardware, keeping the selected workload inside that environment.
- Frontier: task context sent to a supported external AI leaves the local environment and is then subject to that provider, configuration and contract.
- Hybrid: local and supported frontier intelligence can be used for different workloads while ARKTOR remains the control and execution layer.
Model choice does not grant system access.
ARKTOR keeps the requested scope and available capabilities separate from the AI model. Platform permissions still apply, and sensitive actions can require explicit approval.
ARKTOR complements the security stack.
It can assist with inventory, configuration checks, change investigation, approved remediation and result verification. Endpoint protection, patching, network controls and normal security operations remain necessary.
Durable memory stays deliberate.
ARKTOR Vault keeps approved context locally with source information and user control attached. The current alpha does not yet claim completed encryption-at-rest or multi-user authorisation.
sclabs.uk does not track visitors.
No visitor analytics, advertising trackers or cookie-based profiling are used on this site. Pilot enquiries use ordinary email rather than a hidden lead-tracking form.
Local-first supports privacy-conscious deployment. Compliance still depends on how it is used.
GDPR compliance depends on the actual deployment, data, legal basis, retention, access controls, provider contracts and organisational process. Installing ARKTOR by itself does not make an organisation GDPR-compliant.
Public claims stay tied to tested boundaries.
SC LABS separates engineering proof, product maturity and release readiness. Current physical and software evidence is published in Agent Lab, including the limits of each accepted test.
Bring one real workflow, its data boundary and the result you need. The pilot is scoped before execution.