Fast cloud-native block storage provider Simplyblock has developed its Vela software to facilitate instant Postgres database branches without the need for data copying, leveraging its robust block storage capabilities.
Innovative Software Development
Having successfully created Kubernetes-integrated block storage, Simplyblock has now advanced its offerings by producing system application software that harnesses this technology. CEO and co-founder Rob Pankow believes that Postgres is emerging as the go-to database for AI applications. Notably, companies like Databricks are innovating in this space with products such as Lakebase, a serverless Postgres database tailored for AI agents. Additionally, the acquisition of Neon for billion in May 2025 underscores the growing importance of serverless Postgres technology. Snowflake has also made strategic moves, acquiring Crunchy to enhance its Postgres capabilities, positioning itself within the AI Data Cloud. Pankow asserts that Postgres is becoming the cornerstone of contemporary AI development, and Simplyblock aims to capitalize on this trend.
Pankow elaborates, “Vela brings together the speed of serverless computing, the structure of version control, and the resilience of modern storage into a cohesive database platform. By integrating orchestration, virtualization, and high-performance storage under a single API-driven architecture, we empower teams to operate Postgres as a modern platform, transcending the traditional database model.”
Challenges with Managed Postgres Instances
Despite the advantages of managed Postgres instances, Pankow highlights several challenges faced by engineering teams. “I engage with these teams weekly, and they consistently report similar issues. As they attempt to scale a managed Postgres instance during model rollouts, they encounter IOPS limits, throttling windows, and latency spikes precisely when they require reliability. Moreover, costs can escalate dramatically, as the only way to ensure safety is through overprovisioning. These challenges often start small but become unmanageable as AI workloads reach production scale.”
Managed databases tend to offer a one-size-fits-all solution, which is insufficient for the demands of AI workloads that require rapid storage and predictable performance. Additionally, the need for large and frequent database clones for testing and experimentation poses significant hurdles. Managed systems often suffer from internal storage bottlenecks, and their cloning mechanisms rely on slow snapshot-restore cycles or full physical copies, both of which can be costly at scale.
Empowering Developers with Vela
To address these issues, developers are encouraged to run Postgres within their own cloud accounts, co-locating data and GPUs. Vela enables the deployment of Postgres on the same instance as the storage, utilizing the speed and performance of local NVMe devices. This approach ensures resilience and scalability while offering copy-on-write functionality, which is typically unavailable with local storage managed within the user’s cloud instance.
A database branch, as defined by Vela, is a clone that avoids the traditional drawbacks of database copying by employing a copy-on-write mechanism. This concept builds upon the cloning idea but incorporates Git-like semantics and lifecycle management. Each branch is a named, versioned line of development derived from a parent branch, organized in a hierarchical structure. Users can easily diff changes between branches, merge or promote changes back to the parent, rebase, or resolve conflicts.
Cloud-Agnostic Architecture
Vela is designed to be self-hostable and cloud-agnostic, featuring deep integration with Kubernetes. It introduces Git-style branching and facilitates instant environment creation. Vela manages autoscaling, snapshots, and lifecycle orchestration natively, while also incorporating integrated authentication, row-level security, a self-service UI, and observability within the control plane. For high-performance needs, it supports RDMA networking via RoCE alongside NVMe/TCP.
To further support developers, Simplyblock has launched a hosted Vela Sandbox, allowing users to explore the branching model and quickly spin up full environments with minimal setup. Once developers are prepared to scale, Vela can operate within their infrastructure—whether on Kubernetes, bare metal, or virtual machines—ensuring data sovereignty, security, and operational control.
A Vela Beta is currently available, complete with documentation, guides, and access to the hosted sandbox here.
