Single-VM Deployment
An architecture where an entire application, including all its services, runs on one virtual machine.
A single-VM deployment is an architecture in which an entire application or platform, including all of its constituent services, runs on one virtual machine or comparable single server, rather than being distributed across a cluster of multiple machines.
How it works
On a single VM, individual services are usually isolated from one another using containers, each running as a separate process with its own filesystem and resource limits, while sharing the same underlying kernel, CPU, memory pool, and network interface. Orchestration in this model is typically handled by a tool like Docker Compose rather than a distributed scheduler, since there is only one machine to place workloads on. A reverse proxy commonly sits in front of the services to route incoming requests and terminate TLS.
Single VM vs Kubernetes cluster
A Kubernetes cluster distributes workloads across many machines, giving it higher ceiling capacity, built-in failover if a node dies, and the ability to schedule work dynamically based on resource availability, at the cost of meaningfully more operational complexity, since cluster networking, node management, and Kubernetes-specific tooling all have to be learned and maintained. A single-VM deployment trades that headroom and redundancy for a dramatically simpler operating model: one machine to patch, one place to look for logs, and infrastructure that a small team can understand and operate without dedicated platform engineering expertise. For workloads that do not need multi-node scale, this is often a more practical starting point, and sometimes an adequate long-term one.
Why it fits self-hosted AI agent platforms
Running a fleet of AI agents does not inherently require distributed infrastructure. Agent containers are relatively lightweight compared to, for example, a distributed data processing job, and a single reasonably provisioned VM can run a control plane, supporting services, and a meaningful number of agent containers concurrently. For teams evaluating self-hosted agent infrastructure, a single-VM deployment path lowers the barrier to trying and operating the platform, since it avoids the setup and expertise a Kubernetes cluster demands.
Agenhood's implementation
Agenhood is designed to run its full stack, the control plane, connectors service, agent containers, egress proxy, and SearXNG, on a single VM using Docker Compose, with Traefik handling TLS and reverse proxying in production, and an optional Coolify-based deploy path. No Kubernetes is required to operate it.