Cloud Infrastructure Evangelist

Versaia runs its agent orchestration platform on Northflank with 60% lower cost. Tl;dr. Versaia builds an agent orchestration platform for customer and employee experience across voice, messaging, WhatsApp, and chat. They serve enterprises, healthcare, and local government in the Netherlands and the EU. Vahid Vafaei, Co-founder and CTO at Versaia, moved the company off AWS onto Northflank after evaluating Heroku, Fly, Render, and Railway. The full migration took less than two weeks. Staging was stood up in a Sunday afternoon. Two numbers that came out of the move: * Voice engine throughput went from 4-5 concurrent calls per node to 15 concurrent calls per node, roughly a 4x improvement. * Compute cost for their main workload dropped 60%, with no infrastructure to manage. They run their full stack on Northflank today: Go and Python backend services, a React frontend, MongoDB and Redis as managed add-ons, self-hosted Loki, Tempo, Grafana, Qdrant and Dgraph. About Versaia. Versaia builds and orchestrates AI agents that handle customer service and internal employee service for organisations of all sizes. The agents work across voice, messaging, WhatsApp, and chat. Their focus is the EU market, with strong traction in Dutch enterprise, healthcare, and local government. Before Northflank: AWS. Versaia ran on AWS in the Frankfurt region which meant a lot of overhead for a product team that wanted to spend its time on agent orchestration, not infrastructure configuration. "It took a lot of work for us to maintain it. It took a lot of work for us to improve it." They evaluated Heroku, Fly, Render, Railway, and Northflank. Through a ChatGPT search, I ran into Northflank. Northflank Ltd. gave it a try, and since the first time, it was a match made in heaven on how Northflank Ltd. would like to do infrastructure at Versaia. Why Northflank won. A few things mattered more than the marketing claims of any one platform. Deployment seamlessly worked. Render presented itself as plug-and-play but didn't behave that way in practice. With Northflank, deploys went through on the first attempt. "With other platforms, getting an application deployed was a pain. They present themselves as plug-and-play, just connect your repositories, all good to go. It didn't work like that. With Northflank, you get up and running really fast. You just have to have a Docker image. You connect your repositories. I never had anything with Northflank that took two, three tries. It always went." Build and deploy are separate. Northflank decouples builds from deployments and lets you toggle CI/CD on and off per branch. Vahid had recently merged something to main by accident, didn't want it to deploy, and toggled CD off without any issues. The right level of abstraction and developer experience. Vahid has a Kubernetes background and a software engineering background. Northflank sat in the middle. "The UI is very intuitive compared to other solutions. As a company, you always have to find the balance between self-service and how you abstract things. Northflank really has found the right balance there. Especially for someone like me. I have a Kubernetes background, but I have a software engineering background. And this is for me really right there in the middle, the sweet spot." The migration. Everything took two weeks from decision to fully moved over. "We made the decision quite fast. Eventually we moved within two weeks the entire infrastructure." The stack on Northflank today. Everything runs on Northflank: * Backend services: Go and Python * Frontend: React * Database: MongoDB (Northflank managed add-on) * Cache and in-memory work: Redis (Northflank managed add-on) * Object storage: MinIO on Northflank for recordings and transcripts, plus an external bucket * Observability: Self-hosted Loki, Tempo, and Grafana * Store: Self-hosted Dgraph and Qdrant They make use of both Nortrhflank's UI for infrastructure management and the API. Results. Voice agent throughput, 4x. After a CPU upgrade on the Northflank side, the voice engine went from processing 4-5 concurrent calls per node to 15. Compute cost down ~60%. Scale-down is fast. Versaia specifically called out that Northflank's scale-down behavior is among the fastest they've used. Products like Fly.io require significantly more infrastructure work than Northflank. Support is in the same time zone. "Very fast support, very capable people, always ready to help. The fact that you guys are European also really helps in the same time zone." Sandboxing: every workload is isolated by default. Northflank runs every workload in its own sandboxed environment. Each service, job, and build gets isolated compute, isolated networking, and isolated storage. There's no shared runtime between tenants, no shared filesystem, no shared kernel namespaces where they matter. The platform assumes that any workload could be hostile and structures itself accordingly. For Versaia, this is important because their agents handle data for enterprise, healthcare and public sector customers. The isolation model is what lets them run multi-tenant workloads on shared infrastructure without inheriting the blast radius of a noisy or compromised neighbor. Conclusion. For Versaia, Northflank handles the workloads, the costs are predictable, and the team has stopped thinking about infrastructure as a thing they need to maintain. Support is fast, technically capable, and in the same time zone as Versaia's customers. "The level of support we get from you guys is really a game changer for us to stick around. It gives us really a lot of loyalty points towards you."

Today Northflank, the workload platform for developers, announced $22.3 million in new financing across a $16 million Series A led by Bain Capital Ven