™Your infrastructure
Run LibraOS on-premise or in a private cloud. Place models, knowledge stores, and application services where your requirements allow.
Control where
your AI runs.
Choose where your data lives, which models your app uses,
and where actions need approval.
Deployment controls.
Local when required
For an air-gapped setup, use local models and connectors and validate all network dependencies, logs, and update paths.
Reviewed actions
Configure permissions, approval points, and audit for the tasks your app runs through LibraOS. Review the execution path before launch.
Inspect requests
and responses.
LibraOS provides three tiers of AI firewall screening. Configure them for your models and workflows, then evaluate both detection and task quality.
Screening reduces risk; it does not guarantee that every harmful prompt or response will be detected.
- 1Pattern screening
Rules identify configured injection patterns and sensitive-data formats.
- 2Risk scoring
A classifier assesses requests and responses beyond the pattern checks.
- 3Model review
A judge model evaluates ambiguous cases according to the configured policy.
Where controls apply.
Make your review points, source access, and audit requirements part of the application design.
Native Runtime execution
Tasks running through LibraOS can use its configured permissions, approval points, and audit. Test the behavior against your application’s actual actions.
External tools and frameworks
Routing a model request through LibraOS applies controls at that request boundary. Tools, shell commands, and network activity outside the Runtime require their own enforcement.
Enterprise requirements.
Review data placement, identity, models, tool permissions, logs, and support with our team.
Security review
Review the documented controls, deployment boundary, and operating responsibilities for the service you are evaluating.
Regulated environments
Confirm the agreements, responsibilities, and technical controls required for your regulated use case before deployment.
Canadian deployment
Canadian-jurisdiction deployment is available. Confirm the placement of customer data, model traffic, logs, and knowledge stores within the selected environment.