On a more personal note, some words on why I am working on this: crawshaw.io/blog/building-a-…

Check out exe.dev. You can allocate up to 50 VMs, but they split the cpu usage of your fixed monthly plan. It’s cool. Also their integrations are awesome- access git etc without exposing secrets to the vm.

Why can't more sandboxes be like exe.dev

VMs are far from perfect security, but they are orders of magnitude more secure than containers. Practice safe computing, have your agent wear a VM.

There are many ways to manage memory between multiple agents. Here's an ultra-minimal approach for agents you carefully manually curate.

Agent swarms, /goal, and Astra one-shot games get all the attention, but the thing I appreciate most is leveraging agents to do the same work I've always done, just faster. That doesn't mean that I don't love one-shot prompts for software that is just for me. Lately, I'm enamored with exe.dev for building these sorts of personal one-offs (and I feel as though I owe @davidcrawshaw a public apology for being vaguely dismissive of it when it was announced)

I wanted a personal "box in the cloud" to have (1) an interactive agent on top of my Obsidian notes, and (2) explore coding agents and personal agents. Why "box in the cloud"? So that I can (1) access from personal laptop and mobile phone, (2) have fire-and-forget tasks, (3) long-running agents, (4) periodic tasks, (5) experiment without constraints of laptop such as accidentally closing the lid.

@davidcrawshaw wdyt of trynix.dev/ ? I had this idea to replace the ghostty terminal with a @exedev instance instead. * fixes emulation speed * full network access * makes github.com/marketplace/actio… insane Every PR can spin up an exe.dev instance and immediately test their service live with the power of #nix

p(doom)=1 on any programmable device, given sufficient time

Updated pricing page for exe.dev coming soon: make it easier for businesses to buy compute. Everyone keeps asking, it’s time to codify all the custom deals we keep making.

very big fan of exe.dev!

Wife told me these wouldn’t fit. Little did she know I had trained for this moment for years.

this is a very under hyped product: @exedev you can basically use a prompt to spin up computers with an agent embedded in them you prompt what you want this computer to do for you and it just builds it I am still exploring various use cases but here’s an example from this morning: I wanted to test two different open source projects for video editing feedback (like frame.io) I prompted it to create a vm to set both up, test the two projects for my requirements, the key one that I want my agents to easily access my video feedback. it got back me with both projects fully set up and their URLs for testing + a recommendation for which one is better for my needs another example I recently used it for: I wanted to build a CRM + a hermes agent to control it. I prompted exactly that (something like "build a crm for a construction company based in LA [website link]. also create an embedded hermes-powered agent that can control the crm", then it literally one-shotted the crm + the agent. it uses your chatgpt subscription. but its harness can use lots of different providers and models it's very very cool, great job @davidcrawshaw

The old quote that "computers are useless, they can only give you answers" seems to extend to an LLM solution to a millennium problem. Now we have a blowup in Navier–Stokes equations, but no new insights into mathematical technique or turbulence.

I wrote a deep dive into exe.dev `ssh exe.dev new` and half a second later you have a Linux VM with a persistent disk and a private HTTPS URL You pay for a pool of CPU/RAM, not per VM. A VM is as cheap as a folder! flaviocopes.com/exe-dev/

@exedev is absolutely fantastic. I don’t even run my coding sessions locally. I use @herdrdev to remote into my exe.dev machine and can switch devices, close my laptop etc and continue exactly where I was. Game changer

Really liking exe.dev Mine is running my: - Hermes bot - Executor.sh MCP server It comes with an LLM that can access the box. So you can just ask it thinks like 'how secure is this VM from the public?' And then it will fix it for you 🪄

What happens to software when it's not just built by agents, but used and integrated by them too? When the integrator is an agent, does buy still beat build? @svale sits down with @davidcrawshaw (@exedev) and @tidocarriero (@cursor_ai) at Everything *[NYC] 2026 to dig into what that shift means for the infrastructure underneath it all. Sept 9 at Everything [NYC]

Finally got sick of all the Reddit UI changes, so started an @exedev VM and asked it to create an account and make a scraper for the content I check. Now I have an ultra lightweight, dense, mobile UI.

This is why exe VMs run on NVMe. Time for a blog post on this, I think.

this old tweet is getting some likes (why? mysterious), so I re-read it, and still like it. it also applies really well to what I am working on now: I need lots of VMs for agents, but I don't need lots of CPU/RAM. hence @exedev