Grass - Gives your coding agent a dedicated VM that's ready 24/7
by•
Grass gives your coding agent a dedicated VM that's always ready. No setup, no config, no burning your laptop.
Point Claude Code or OpenCode at it and run. Monitor progress from your phone, steer mid-session, push changes, all without touching your machine.
Every new account gets 10 hours free. No credit card needed.
Replies
brag.fast
Congratulations on the launch. This looks like something I’d use. I use my agents from different places and it’d be cool to keep on developing from my mobile for example.
Grass
@rob_vb Thanks Rob! That's exactly the use case. Pick up where you left off, from your phone, without losing any context. Good luck with @brag.fast today too.
Hello!
Does this work if I'm not on wifi? Like can I actually check in on my agent from anywhere?
Grass
@shubham_mandge Yep, works over mobile data too!! The VM runs on our end so your phone just needs any internet connection. No laptop, no wifi dependency
The BYOK architecture is the detail that matters most to me. I've avoided cloud agent tools specifically because I didn't want my API key sitting on someone's server. Good to see this addressed upfront rather than buried in the docs.
Grass
@jay_morzaria This is exactly why we built it that way. Your key, your model access, your bill. We never see it, never store it. Felt like the only honest way to build something in this space.
Okan
Running a local coding agent usually ends up chewing through all my RAM and crashing my active dev environment. Spinning up a persistent VM specifically for this is a brilliant way to handle long-running background refactors. I would love to know if you offer pre-configured images for different agent frameworks out of the box.
Grass
@y_taka Local agents can get overwhelmed, especially for long-running tasks.
Right now, we offer pre-configured environments for Claude Code and OpenCode out of the box, with Codex and Pi coming next.
We’re also thinking about this more as ‘agent-ready workspaces’ rather than just base images. So the goal is to have setups that are tuned for specific workflows/frameworks over time.
Curious, what coding harness or setup are you using today?
SaveMRR
The environment setup tax on coding agents is real. I waste so much time getting the right dependencies installed before the agent can actually do anything. How fast is the cold start?
Grass
@kailesk_khumar We see cold starts in the ~10–20 second range right now.
The goal is to keep it fast enough that you don’t feel the setup tax. Spin up, connect your agent, and get straight to work without dealing with dependencies. Auth for Claude code takes another 20-30 seconds 😅 (it's a one time thing).
The "no burning your laptop" part sold me haha. Being able to just point Claude Code at a remote VM and watch it work from my phone sounds like the dream setup. How does the git integration work -- does it push directly to my repo or do I review a PR first?
Grass
@antoninkus Right now, we don’t enforce a specific Git workflow. Once you connect your GitHub account, the agent can use git directly via bash. So it can commit, push, or open PRs depending on how you want to structure things.
We’re thinking about adding more opinionated flows (like PR-first by default), but for now it’s fully flexible.
What’s your preferred workflow with agents? Direct pushes or PR review first?
Hi, I'm an OpenClaw user and it looks like a very convenient version of OpenClaw to me.
BTW how can I transfer files to and from the VM?
Grass
@seunghoon_lee4 Nice, love OpenClaw :)
Right now, the main way to move files is via Git. Since your repo is connected, the agent can pull/push changes directly as part of its workflow.
You can also use standard tools inside the VM (like curl, wget, etc.) if needed.
Cool idea but whats the plan when a session just keeps going and burns through hours? Some kind of cap would be nice. Also what specs do the VMs have, Claude Code gets pretty hungry on larger projects ...
Grass
@daniel_punke Thanks! For the managed version, we currently offer ~10 hours free as a trial, and we’re planning to introduce usage caps + controls so sessions don’t accidentally run forever.
On specs: each workspace runs in a container with a couple of GB of RAM and ~5GB of storage right now—we’re still tuning this based on real workloads (especially for heavier agents like Claude Code). Not sure what we'll land on just yet.
Also, the Grass server is open source, so you can run it on your own VM with whatever specs you need, no limits.