Zero-trust baseline
No layer gets trusted by default. Browser, API, database, CI, AI model endpoints, and cloud services are authenticated, scoped, and logged where the stack supports it.
Security
Security is not an afterthought.
Security is no joke. Every Entoura production build starts from a zero-trust baseline: never trust, always verify, and give every person, service, key, and system the least access it needs to do the job.
Operating principle
Never trust. Always verify. Least privilege everywhere.
No boundary is casual. Browser to API, API to database, CI to cloud, and contractor to repository are treated as security decisions, not setup details.
Least privilege
Accounts, service keys, roles, and contractors get the access required for the task. Nothing more. Access is reviewed before handoff and after offboarding.
Secrets discipline
Production secrets and API keys (including model provider keys) live in encrypted platform stores, not source code, logs, build output, Slack, or email. Service-role keys never ship to the client.
Data isolation
Database access is designed around explicit policies, private storage by default, signed URLs where needed, and organization-level boundaries.
AI infrastructure security
Model access controlled per environment. Data sent to inference endpoints is defined in writing during the Blueprint: what reaches a model, what stays local, what gets logged. Prompt injection and output validation are handled at the application boundary. Inference residency is a deployment decision, not an afterthought.
Client-controlled setup
Source code access, hosting, domains, keys, billing, deployment access, and documentation are planned so clients have practical control where platform rules allow.
Expo application services
Mobile builds use a platform with enterprise security controls.
When a mobile app is built with Expo and EAS, the build and distribution pipeline benefits from Expo's documented security and compliance program. This supports the project; it does not replace application-level security, privacy scoping, or client-specific compliance work.
SOC 2 Type 2
Expo states that Expo Application Services is SOC 2 Type 2-compliant for the Security trust services criterion.
Encrypted data
Expo documents encryption in transit and at rest, including HTTPS in transit and AES-256 or stronger encryption at rest.
Build isolation
EAS Build workers are ephemeral virtual machines cleared after each use, with build logs and artifacts retained for limited windows.
Cloud infrastructure
Expo services are primarily hosted on Google Cloud Platform, using GCP physical and logical infrastructure security.
Access controls
Expo supports MFA, enterprise SSO, and audit logging for administrative activity where the account plan supports it.
Privacy posture
Expo documents GDPR, CCPA, and Data Privacy Framework compliance for the user data it processes.
AI supply chain
A model you call is a vendor in your stack. We treat it like one.
The AI ecosystem is wide open. Anyone can pull a tool, plugin, agent, or repository off the internet and wire it into a workflow, and a lot of it is unsafe: prompt injection, data exfiltration, poisoned dependencies, the kind of thing that quietly turns your own systems against you. Most businesses stitching these parts together have no way to know what they've let in. Every component we build with is vetted, verified, and hardened before it goes anywhere near your data, so you get the capability without inheriting the risk.
Model provider as a dependency
A model you call is a vendor in your stack. Its endpoint, terms, uptime, and data handling are part of your supply chain, so we scope them in the Blueprint the same way we'd scope any critical dependency.
What reaches a model, what stays local
The data boundary is defined in writing before the build: what gets sent to inference, what stays on your side, and where inference runs. Residency is a deployment decision, settled up front.
Prompt injection and poisoned inputs
Untrusted content reaching a model is an attack surface, the route to prompt injection, data exfiltration, and poisoned inputs. We handle it at the application boundary, with output validation, so capability doesn't arrive with a hidden cost.
Production basics
What ships with every serious build.
Security is part of the scope, not polish added at the end.
- Authentication and role-based permission structure
- Protected API routes with validated inputs
- Secrets kept out of source code and client bundles
- Private-by-default storage and signed access where needed
- AI model access scoped per environment with key rotation
- Prompt injection safeguards and output validation
- Inference data residency documented before build
- Preview and production deployment flow
- Repository, hosting, account, and credential handoff planning
- Operator documentation and future-developer notes
- Incident response path for suspected compromise
Regulated work
Specific requirements get scoped explicitly.
Canadian hosting regions, Expo/EAS security posture, privacy notices, audit needs, vendor reviews, backups, retention, and procurement questions are handled during scoping. Entoura can support the documentation and technical conversations, but does not claim blanket certification for every project.
Discuss requirements →