Model Library Reference Manual v1.3
The Model Library serves as the central governance system for AI model images within EDB Postgres AI, providing organizational control over model lifecycles while enabling integration across Pipelines, Model Serving, Knowledge Bases, and Gen AI Builder.
Prerequisites: The Model Library requires Hybrid Manager Asset Library installation. All model metadata, versioning, and lifecycle operations are managed through the Asset Library infrastructure.
Purpose and Function
Organizations implementing AI at scale face fundamental challenges with model management: fragmented repositories, version inconsistencies, compliance gaps, and limited operational visibility. The Model Library addresses these challenges through centralized governance that maintains control without sacrificing development velocity.
The system operates as a curated gateway between external model sources and internal AI capabilities, ensuring only validated models reach production environments while providing comprehensive audit trails for compliance requirements.
Core Architecture
Registry Integration Layer
The Model Library connects to external container registries through configurable synchronization rules. This layer automatically discovers, validates, and ingests model images based on organizational criteria while maintaining security boundaries.
Supported Registry Types:
- Harbor (private container registry)
- Amazon Elastic Container Registry (ECR)
- Azure Container Registry (ACR)
- Google Container Registry (GCR)
- Custom registry implementations
Governance Framework
Built-in approval workflows validate models against organizational policies before production availability. The framework supports complex multi-stakeholder approval processes while maintaining automated validation for standard compliance requirements.
Policy Enforcement Points:
- Security scanning and vulnerability assessment
- Performance benchmarking against organizational standards
- Compliance validation for regulatory requirements
- Digital signature verification for model provenance
Distribution Hub
Approved models become available across all EDB PG AI capabilities through standardized interfaces. This ensures consistent model versions across development, staging, and production environments while supporting environment-specific configurations.
Operational Workflows
Model Registration Process
Automated Discovery Repository rules define which external sources to monitor for new model versions. The system continuously scans configured registries and automatically ingests models matching defined criteria, reducing manual overhead while ensuring policy compliance.
Validation Pipeline Newly ingested models undergo automated validation including security scanning, metadata extraction, and performance verification. Models failing validation checks are quarantined until issues are resolved or explicitly approved through override procedures.
Approval Routing Validated models enter approval workflows based on classification and risk assessment. The system routes models to appropriate stakeholders while providing automated approval for models meeting predefined criteria.
Lifecycle Management
Version Control The system maintains immutable version history for all registered models, supporting semantic versioning schemes and backward compatibility validation. Version tracking enables rollback capabilities and dependency management across complex model ecosystems.
Promotion Pipelines Models advance through environment tiers (development → staging → production) based on validation checkpoints and approval gates. Automated promotion reduces deployment friction while maintaining quality control.
Deprecation Handling Configurable retention policies manage model lifecycle transitions, providing advance notification before deprecation and ensuring safe retirement without disrupting active deployments.
Integration Points
Model Serving Platform
The Model Library provides the curated catalog for Model Serving deployments. Only models approved through the governance framework can be deployed to production inference endpoints, ensuring consistent security and compliance standards.
Integration Benefits:
- Automated model updates with validation checkpoints
- Rollback capabilities for problematic deployments
- Performance metadata optimization for resource allocation
- Seamless KServe infrastructure integration
AI Accelerator Pipelines
Pipeline execution leverages validated models from the library, ensuring consistency across AI Accelerator Pipelines stages and environments. The system provides performance characteristics that optimize pipeline resource allocation and execution strategies.
Gen AI Builder
Gen AI Builder applications access governed model collections through the library interface. This ensures knowledge base implementations use validated embedding models and assistant creation references approved language models.
Security and Access Control
Multi-Tenant Architecture
Role-based access control aligns with organizational hierarchies while providing namespace isolation for different teams and projects. Fine-grained permissions control model registration, approval, and deployment activities based on user roles and organizational policies.
Security Validation
Automated vulnerability scanning examines all registered model images for known security issues. The system validates model signatures and provenance information during registration while implementing policy-based controls that prevent risky deployments.
Security Features:
- Digital signature verification for model authenticity
- Vulnerability database integration for threat detection
- Policy engine for automated security control enforcement
- Comprehensive audit logging for security monitoring
Implementation Considerations
Registry Configuration
Establishing registry connections requires careful consideration of network topology, authentication mechanisms, and synchronization frequency. Organizations must balance automation benefits with security requirements when configuring repository rules.
Critical Factors:
- Network connectivity and firewall requirements
- Authentication and authorization mechanisms
- Synchronization frequency and resource impact
- Error handling and retry strategies
Configure Private Registry Integration →
Approval Workflow Design
Approval processes must balance governance requirements with development velocity. Organizations should design workflows that automate routine approvals while ensuring appropriate oversight for high-risk models.
Design Principles:
- Risk-based routing for efficient stakeholder involvement
- Automated approval for models meeting standard criteria
- Clear escalation paths for complex approval scenarios
- Comprehensive audit trails for compliance requirements
Metadata Management
Comprehensive metadata capture supports discovery, governance, and operational requirements. Organizations should establish consistent tagging schemes and documentation standards that align with business requirements.
Operational Patterns
Enterprise Governance
Large organizations benefit from the Model Library's sophisticated approval workflows and comprehensive audit capabilities. The system supports complex multi-stakeholder processes while maintaining development velocity through intelligent automation.
Typical Implementation:
- Development teams register models in staging environments
- Automated validation verifies security and compliance requirements
- Risk-based routing directs models to appropriate approval stakeholders
- Approved models propagate across all EDB PG AI capabilities
- Production deployments automatically reference validated versions
Sovereign AI Deployments
Organizations requiring complete model sovereignty leverage private registry integration and air-gapped deployment capabilities. This pattern ensures sensitive models never leave organizational boundaries while maintaining enterprise operational capabilities.
Key Characteristics:
- Complete air-gap support with offline distribution
- Private registry integration with metadata preservation
- Automated compliance reporting for regulatory requirements
- Classified model handling procedures and controls
Development Consistency
Teams benefit from consistent model availability across development lifecycles. The Model Library eliminates version drift between environments while supporting environment-specific configurations and deployment strategies.
Limitations and Considerations
Performance Impact
Registry synchronization and model validation processes consume computational resources. Organizations must consider infrastructure capacity when configuring synchronization frequency and validation depth.
Network Dependencies
The system requires reliable network connectivity to external registries for synchronization operations. Network failures or connectivity issues can impact model availability and synchronization schedules.
Storage Requirements
Model image storage scales with organizational model inventory. Large organizations with extensive model catalogs should plan for significant storage requirements and associated backup strategies.
Getting Started
Initial Setup
- Verify Hybrid Manager installation and configuration
- Establish network connectivity to required external registries
- Configure initial repository rules for model discovery
- Implement basic approval workflows aligned with organizational requirements
Configuration Resources
Best Practices
- Implement least-privilege access controls for all model operations
- Establish regular security scanning schedules for registered models
- Maintain comprehensive backup and disaster recovery procedures
- Document approval workflows and escalation procedures clearly
Reference Documentation
Core Components:
Integration Capabilities:
Implementation Guides:
The Model Library transforms fragmented model management into systematic governance that scales with organizational AI initiatives while maintaining security, compliance, and operational excellence across all EDB Postgres AI capabilities.