Legal
Security
Last updated: 2026-07-27
Local storage by default
JARVIS runs as a native macOS application. Conversation history, tasks, cached context, and credentials for connected services stay on the user's device: product data is held in a local SQLite database, and API keys are held in the macOS Keychain. JARVIS AI operates no sync service and no product backend, so none of this data is uploaded to us. Where model inference runs is a separate question, answered below.
Planned AWS cloud control plane
The shipped product has no cloud sync, so no product data reaches JARVIS AI infrastructure today. Optional cloud features are planned on AWS in the Asia-Pacific (Singapore) region and would use Amazon Cognito for authentication, AWS KMS customer-managed keys (CMK) for envelope encryption, S3 with bucket-level encryption and TLS-only access for sync data, and DynamoDB for user metadata. Those stacks are defined as code (AWS CDK) and reviewed against the AWS Well-Architected Framework, but they are not connected to the application. This page will be updated before any of them handles customer data.
Encryption at rest and in transit
The only store of product data today is the local SQLite database on the user's device, protected by macOS file permissions and by whatever full-disk encryption the user has enabled. API keys and service credentials are kept in the macOS Keychain rather than in that database. Network traffic to this website and to third-party APIs uses TLS 1.2 or higher. Envelope encryption with a per-user data key wrapped by AWS KMS is the design for synced content once cloud sync ships; it is not in production, and there is no cloud-stored customer data to which it applies.
Where model inference runs
Inference is not local by default. On a standard install JARVIS sends the request to a cloud model provider, currently Google's Gemini API, so Google processes the message text and any email, calendar, or note content included in that request. The provider API key is supplied by the user; no model credentials are bundled with the software. Local inference is supported and can be enabled in Settings under Local AI, which keeps the request on the device. Amazon Bedrock is used only when the user configures AWS credentials and selects it for heavier inference. Prompts and completions are stored only in the local database on the user's device; JARVIS AI does not receive or retain them.
Audit logging and monitoring
No customer data lives in JARVIS AI cloud infrastructure today, so monitoring currently covers our own operations: AWS CloudTrail event history records control-plane actions in our accounts, and billing alerts flag anomalous spend. Fuller observability commitments will be published here before any cloud feature handles customer data.
Service providers
Model requests are processed by the provider configured in the application. On a standard install that provider is Google: its Gemini API receives the message text and any email, calendar, or note content the assistant is asked to work with. If the user enables a local model, those requests stay on the device instead. Website form submissions are delivered through Resend, a transactional email provider. Payments are processed by Stripe; JARVIS AI never stores raw card numbers. Each provider is bound by its own published security and data processing commitments.
Access controls
JARVIS AI is a single-operator company. AWS access is limited to scoped IAM identities used for release and infrastructure work, and there is no customer data in those accounts for any identity to reach. Formal role separation will be introduced, and documented here, before cloud features handle customer data.
Reporting a security concern
If you believe you have found a security issue with the website or the product, please email contact@jarvis-automation.com with [Security] in the subject line. We aim to acknowledge reports within 48 hours and keep reporters informed until the issue is resolved.