.png)
How ProtectAI Scaled Secure On-Prem AI Without Slowing Engineering
“A lot of companies spend years building bespoke delivery systems. Replicated lets us move faster while staying aligned with standards that are maintainable long-term.”
.png)
Overview
ProtectAI provides security solutions for AI and machine learning workloads across their entire lifecycle. Its customers include enterprises operating in highly regulated industries such as banking, insurance, and government, many of which require software to run in customer-controlled or fully air-gapped environments
Anticipating these constraints, ProtectAI designed its platform from the outset around a single-tenant model for self-hosted and on-prem deployments. To support that approach at scale, ProtectAI uses Replicated to deliver its platform into restricted environments without pulling engineering focus away from core AI security innovation.
“We have customers that won’t accept a SaaS component… and at least for our company, that represents a lot of our contract dollars.”
Challenge
“Some customers, especially in government or air-gapped environments, don’t have a choice,” Brookins said. “On-prem isn’t optional.”
Even when SaaS deployments are technically allowed, approval cycles can take six to eight months. In contrast, deploying software directly into a customer’s environment, whether on-premises, in a VPC, or fully air-gapped, can dramatically reduce time to value while building trust with security-conscious buyers.
Supporting these environments wasn’t optional for ProtectAI, but it introduced a different challenge. Delivering on-prem software at scale requires installers, registries, upgrade paths, and support tooling, all of which can quickly consume engineering time.
“We’re interested in building our product, not tools that help our product,” Brookins said. “Once you build bespoke distribution tooling, you also have to maintain it forever.”
Solution
ProtectAI chose Replicated as the foundation for distributing its platform into customer-controlled environments, resolving the need to build and maintain custom distribution tooling in-house.
“As a startup, I can’t imagine doing this without someone who’s already done a lot of the work,” Brookins said. “A lot of companies spend years building bespoke delivery systems. Replicated let us move faster while staying aligned with standards that are maintainable long-term.”
While Replicated simplified how ProtectAI packaged and delivered its application, securely distributing container images into restricted and air-gapped environments was the next critical challenge to solve.
Guardrails for Standardized Kubernetes Delivery
ProtectAI values Replicated’s opinionated guidance for Kubernetes-based deployments.
“Kubernetes is a wild west,” Brookins said. “Replicated gives you guardrails, guidance on what you should and shouldn’t do, which helps keep things standardized and maintainable.”
Beyond the technology itself, ProtectAI views Replicated as a long-term partner.
“The honeymoon period never went away,” Brookins said. “Replicated has been consistently engaged in a way that honestly makes other vendors look bad sometimes.”
Results
By using Replicated to standardize how its platform is delivered, ProtectAI improved both customer experience and internal engineering efficiency while scaling its on-prem and air-gapped deployments.
As a result, ProtectAI achieved:
- Faster deployments for regulated customers requiring on-prem or air-gapped installations
- Reduced engineering overhead by avoiding custom distribution tooling
- Improved long-term maintainability through standardized, opinionated deployment patterns
- Greater internal velocity, allowing engineers to focus on ProtectAI’s core AI security capabilities
Today, ProtectAI delivers its AI security platform into some of the most restricted environments in the market without slowing engineering teams or compromising on maintainability.
.svg)
.png)