Nadrama is building the compute layer specifically for the agent ecosystem. Agents differ from traditional web applications because they are often long-running, require bursty scaling, and handle highly sensitive data that mandates strict environment control. Nadrama’s Kubernetes PaaS addresses these needs by providing a dedicated, self-hosted environment where agents can execute without the latency or privacy concerns of multi-tenant platforms.
They are active in the "infrastructure and deployment" segment of the agent stack. By offering tools like Netsy (S3-based state) and Nstance (fast scaling), they provide the low-level optimizations necessary for agents to perform reliably. For developers and enterprises, Nadrama matters because it offers a path to move from agent prototypes to production-grade, sovereign infrastructure that works across both public and private clouds.
Nadrama is a 2024 startup building the underlying plumbing for AI agents. Founded by Ryan Djurovich, an engineer with a background at Cloudflare and Xero, the company is tackling the problem of where agents actually live. While the industry has focused heavily on the models themselves, the infrastructure required to run long-running, autonomous agents often defaults to the same proprietary cloud silos that developers have spent a decade trying to escape. Nadrama’s approach is to provide a Kubernetes-based Platform-as-a-Service (PaaS) that users can deploy into their own cloud accounts.
The core of the Nadrama offering is a specialized distribution of Kubernetes designed for the specific demands of agents. Traditional Kubernetes is often criticized for its complexity and the "day two" operations burden it places on small teams. Nadrama attempts to mitigate this by swapping out or simplifying several key components. Their most notable technical departure is the replacement of etcd, the standard Kubernetes state store, with Netsy. Netsy stores cluster state in S3, which simplifies the recovery and scaling of clusters—a critical requirement for agent environments that may need to spin up and down rapidly based on task volume.
Nadrama is not just a single product but a collection of open-source tools that form a cohesive infrastructure layer. This includes Nstance for faster auto-scaling and provisioning, Easy OIDC for simplified identity management, and Prebake for cluster imaging via Helm charts. These tools are licensed under Apache 2.0, reflecting a commitment to software liberty that allows users to leave the platform with their infrastructure intact if they choose. This anti-lock-in stance is a direct response to the "walled garden" approach taken by many emerging agent-hosting startups.
Deploying Nadrama typically begins at the CLI. The platform generates Terraform or OpenTofu configurations that developers then apply to their own accounts on AWS, Google Cloud, or Proxmox VE. The inclusion of Proxmox is particularly interesting. It suggests a focus on users who want to run agents on local hardware or private clouds for reasons of cost or extreme data privacy. Once the infrastructure is provisioned, developers interact with it using standard tools like kubectl, making it a familiar environment for those already experienced with container orchestration.
In the competitive landscape, Nadrama sits between high-level agent orchestration frameworks and raw cloud providers. While frameworks like LangChain or CrewAI handle the "logic" of the agent, Nadrama handles the "hosting." By focusing on the infrastructure layer, they avoid competing with the rapidly shifting world of agentic architectures, instead providing the reliable compute that all those architectures eventually require.
The company is currently in a pre-launch phase, with several of its underlying tools already available on GitHub. Their business model appears centered on a fixed-fee monthly subscription, providing price predictability for developers who are wary of the variable costs associated with traditional cloud providers. As agents become more complex and require more persistent, secure environments, the demand for a dedicated agent-operating-system equivalent will likely grow. Nadrama is betting that this operating system should be open, portable, and under the user's direct control.
Agent infrastructure you control, built on Open Source Kubernetes.
Nadrama is hiring