Controls you can review.
OctOpus processes data, runs code and calls external services to complete your work. This page describes the implemented controls and the limits that depend on your deployment.
Where processing happens
Desktop
The desktop application can run a local backend and save local work. Model calls, connected sources, model downloads, account services and configured remote compute can still use the network. Review those settings and use operating-system disk encryption where appropriate.
Private deployment
Contact us to scope a private or VPC installation. Residency depends on the agreed storage, inference endpoints, network rules and enabled integrations. Verify the actual deployment before relying on a perimeter or region restriction.
Hosted service
The web service uses external hosting, database, storage, compute and configured AI providers. A workflow may send context, samples or full training data to the service that executes it. See the provider and data-handling details.
Encryption and credentials
- Connections. The public site uses HTTPS. Review internal connections and external endpoints for any customer-managed installation.
- Saved credentials. Provider and connector credentials use application-level encryption. Authorized features decrypt them when needed; ordinary key-management responses do not return the saved plaintext.
- Files and datasets. At-rest protection depends on the storage service and deployment. Local files require the device's access controls and disk-encryption settings.
Identity, access and sharing
- Server-side permissions. Account, workspace and project checks govern access. Shared-project roles can permit owners, editors or viewers to read or change work; client-supplied roles alone do not grant access.
- Seat licences. Enterprise domain-based seat claims require a verified company email address and an available seat.
- Shared outputs. Published apps, shared projects and authorized download links have their own access rules. Review the audience before publishing or sharing.
Code execution and logs
- Experiment controls. Runners restrict the environment and credentials provided to generated code. Remote execution and local execution have different isolation boundaries; inspect generated code and configure the appropriate executor for your data.
- Secret redaction. Application log paths mask recognized credentials and secret patterns. Redaction is a protective measure, not a guarantee that every sensitive value will be detected.
- Storage checks. Production startup validates required storage configuration. The ephemeral-host path checks external database and object-storage requirements. Backups, recovery checks and deployment review remain necessary.
Model evaluation and audit records
- Inspect the available experiment code, logs, metrics and saved artifacts in the workspace.
- Review the evaluation split, data leakage checks and task-specific acceptance results. A validation score does not establish production suitability.
- Preserve the dataset version, code, dependencies and configuration for reproduction. Randomness, external models and hardware can affect repeated results.
- Audit coverage and retention vary by workflow. Confirm the records and exports required for your organization's review.
Compliance review
This page does not claim SOC 2 certification or HIPAA or PCI DSS compliance. A private deployment is not a substitute for the agreements, configuration and evidence required for your use case.
Before submitting regulated data, agree the deployment scope, providers, processing locations, retention, access controls and contractual responsibilities with us. Contact ceo@octoopus.dev to request the information available for that review.
Responsible disclosure
Report suspected security issues to security@octoopus.dev. Include steps to reproduce and the affected feature; do not send passwords, secret keys or other people's private data. We will coordinate investigation and remediation with you.