I spent most of this morning trying exe.dev, a platform that offers an interesting alternative to renting and managing a traditional cloud server.
These are just my initial impressions, but the experience left me wondering whether the platform solves a technical problem or simply makes an existing workflow much more convenient.
An abstraction over cloud servers
With a traditional server, I might rent a fairly powerful machine and use Docker containers to separate different workloads. I’d be responsible for configuring the server, managing the containers, and deciding how they share the available resources.
exe.dev abstracts away much of that work. With one account, I can create virtual machines whenever I need an isolated environment for a deployment, an AI agent, or another cloud workload. The current limit is 50 VMs.
That’s an interesting idea, especially for AI-assisted development. Instead of configuring a VM by hand, I can describe the environment I want in natural language.
Provisioning a development environment with a prompt
The interface is minimal. It gives me a prompt where I can explain what I want the VM to do.
For my first machine, I asked for a development environment with Neovim and mise. I also requested the languages I regularly use, including Ruby, Python, Go, Java, and Rust.
Once I finished the description, the platform used an LLM-powered provisioner to configure everything. It did a surprisingly good job. Watching the environment come together from a natural-language description was genuinely impressive.
When provisioning was complete, I received SSH access. From then on, it behaved like a regular remote VM that I could use for almost anything.
The connection felt a bit slow, although the Wi-Fi where I was testing wasn’t particularly good. I need to try it again from home before deciding whether the issue was with the service or my connection.
A natural fit for AI agents
One of the most compelling possibilities is creating several VMs for different agents. I could leave them working on separate tasks and later SSH back into their environments.
Unlike a typical serverless environment that shuts down or loses its local state, these VMs have persistent storage. Files and sessions remain available, making the platform a better fit for long-running agent workflows. An agent can keep working in a server session instead of relying on an ephemeral environment.
The platform also includes a simple web-based agent interface. I can prompt it to perform tasks or install software without doing everything manually through SSH.
The pricing question
The plan I tried costs $20 and includes 8 GB of RAM shared across the VMs. Being able to create up to 50 VMs sounds generous, but I can’t imagine actively using that many with the available memory.
This is where I start comparing the service with running my own server. For around $30 to $35, I could rent a Hetzner server with roughly 32 GB of RAM and persistent storage. I could create my own Docker containers, connect them through Tailscale, and access each environment over SSH. The containers could share the server’s resources, and I could run agents inside them too.
Of course, I’d also be responsible for building and maintaining the whole setup.
Convenience or a bridge over a knowledge gap?
That’s the tradeoff I’m still trying to understand. Is exe.dev mainly selling convenience to technical users who could build the same infrastructure themselves? Or is it bridging a knowledge gap for people who want several AI-powered environments without learning how to manage servers, containers, networking, and provisioning?
The minimal interface and natural-language setup make the platform approachable. At the same time, the documentation seems aimed at technical users. I’m still not sure who the ideal customer is.
For now, I see the appeal. The provisioning experience is polished, persistent VMs are useful, and creating separate environments for multiple agents could enable some interesting workflows. I just need more time with it to decide whether that convenience is worth it compared with managing my own server.