# CoreWeave Registry: The System of Record for Enterprise AI Assets

## Executive summary

CoreWeave Registry serves as a centralized, governed system of record for managing the entire lifecycle of durable AI assets, including models, datasets, evaluations, and LoRAs. It allows teams to track asset lineage, manage versions, and automate deployment workflows, effectively acting as a 'stage gate' for AI assets before production use.

## Key takeaways

- Centralized Asset Management: Registry is designed to manage any durable AI asset (models, datasets, LoRAs, etc.) that needs organization-wide collaboration and reuse. Assets can be imported from external sources like Hugging Face or linked directly from other Forge products, such as Weights & Biases Models (1:49).
- Lineage and Version Control: The Registry automatically establishes relationships, allowing users to view lineage, manage multiple versions, and apply tags and aliases to artifacts, ensuring traceability throughout the AI asset lifecycle (2:00).
- Deployment and Testing: Users can deploy models to an ephemeral testing endpoint directly from the Registry, enabling live requests and management of deployments within the platform (2:20).
- Workflow Automation and Governance: Automation features allow for establishing 'stage gate' processes. This includes setting up alerts (e.g., notifying a Slack channel when a new version is linked) and implementing protected aliases that can trigger webhooks for external CI/CD systems like GitHub Actions (3:43).

## Technical details

- Asset Integration: Assets can be brought into Registry via URL (e.g., Hugging Face) or by linking artifacts from other Forge products (e.g., Weights & Biases Models).
- Collection Cards and Metrics: Users can create dynamic 'Collection cards' (similar to Model Cards) using ARIA. These cards can include live metric panel grids, providing a comprehensive, shareable context for the asset collection (2:54).
- Protected Aliases and Webhooks: Protected aliases restrict who can modify critical production identifiers. Adding a protected alias can trigger a webhook, enabling integration with external automation tools (e.g., GitHub Actions) for advanced CI/CD management (3:43).

## Practical implications

- Build engineers can use Registry's automation and protected alias features to enforce mandatory review and approval steps for AI models before they are promoted to production, formalizing the CI/CD pipeline for ML assets.
- The ability to link lineage and versions directly within the Registry simplifies auditing and compliance requirements for governed AI assets.
- Integrating webhooks allows the Registry to act as a trigger point for external CI/CD pipelines, automating deployment steps upon asset approval.

## Topics

AI Asset Management, MLOps, Data Governance, Workflow Automation, CI/CD, Lineage Tracking, CoreWeave Registry

Source: https://www.youtube.com/watch?v=aj-Yene1dyM
