— Security & reliability
Security is part of the system, not a checkbox.
We design production systems around controlled access, validation, logging, human checkpoints and clear failure states. We do not claim certifications or controls we do not have.
Production systems
A production system requires validation, error handling, retries, monitoring, security, audit trails, integration with real data and tools, testing and support. A prototype is not treated as a production deployment.
Human approval for consequential actions
We do not automate consequential financial, legal or safety decisions without a human approval step. For AI agents, actions are bounded, permissioned and reviewable rather than open-ended.
Data and untrusted website content
Public website content used by the business-analysis service is treated as untrusted data. It is analysed as input and is not treated as instructions to the calling AI system.
Cadence Ops architecture
Cadence Ops is the clearest public example of the production systems we build. Its published architecture includes Postgres with row-level security, a versioned 3D layout model, draft-only AI proposals and human approval before publishing changes.
What we do not claim
We do not claim SOC 2, ISO 27001, penetration-test certification, or other formal security certifications unless and until they are actually in place and documented.
Security questions
If you need to discuss data handling, access controls, deployment, integrations or your organisation's security requirements before a project starts, contact us.