Grass

About

Grass is not trying to replace your coding agent. It replaces the fragile execution setup around it. Instead of keeping Claude Code or OpenCode pinned to one laptop, it gives the agent a persistent machine plus a mobile control surface for approvals, session monitoring, and diff review.

Why It Is Hot Now

It is getting attention because it packages a very real pain point into a simple pitch: your agent should not stop just because your laptop closed. Product Hunt traction and the phone-first approval loop make it feel more concrete than generic cloud-agent infrastructure claims.

Key Features

• Runs agent sessions on a persistent VM so they survive sleep, disconnects, and time away from the desk.
• Sends permission requests to your phone so long-running tasks do not stall waiting for a terminal prompt.
• Lets users review repo-organized diffs, switch between GrassVM and their own machines, and steer sessions remotely.

Real Use Cases

• Starting an overnight bug fix or refactor and checking progress without leaving a laptop awake.
• Approving or rejecting sensitive file writes while away from the desk instead of SSH-ing in just to unblock a run.
• Keeping mobile visibility on agent work for founders or senior engineers who need to supervise multiple repos at once.

Community Pulse

The strongest positive reaction is that Grass attacks the boring infrastructure problem instead of pretending the model alone is the product. The main caution is whether phone-based control is enough for high-stakes engineering review, especially when approvals need richer context than a quick notification can provide.

Limits and Risks

Grass improves continuity and convenience, not agent quality. If the task needs close architectural steering or deep local environment parity, a persistent VM only solves part of the problem. Teams should also evaluate how much trust they are comfortable placing in remote approvals and hosted execution.

Alternatives

Alternatives include Replicas, Runtime, SuperHQ, dedicated self-managed VMs, and simply running Claude Code or OpenCode on a long-lived remote devbox with manual notifications.

FAQ

• Who should evaluate Grass first? Teams or solo developers already using coding agents who mainly want persistent runtime and remote approval rather than a new agent model.
• What should you validate early? Permission UX, repo context on mobile, VM startup and resume behavior, and whether remote review is good enough for your actual merge workflow.