Core concepts
Projects, computers & threads
Four concepts come up everywhere in the product: projects, computers, threads, and agents. Understanding how they relate makes the rest of grasscoding easy.
Projects
A project is a unit of work — typically one repository or one product. Members, billing, integrations (iMessage, Slack, GitHub), and agent configuration all live at the project level. You can have many projects in a single account, and projects can be shared with teammates.
Project settings include workspace details, members, agent prompt configuration, environment variables, preview rules, sandbox snapshots, MCP servers, and skills. Settings persist across every computer in the project, even after you clone or replace the underlying VM.
Computers (sandbox VMs)
A computer is the actual remote machine where the agent operates. Each computer has its own filesystem, its own running processes, its own preview URL, and its own checkout of your repo on whichever branch you point it at. Computers are provisioned via Freestyle and isolated from each other.
You can clone a computer in one click — same files, same env, same install state — to run two agents in parallel on the same project without them stepping on each other. SSH is available if you want to drop in and poke around manually, and a hosted editor and terminal are always one tab away.
Threads
A thread is one conversation with one agent. Threads remember the back-and-forth, the files that were touched, the commands that were run, and the preview link the agent produced. You can fork a thread, replay a turn, or start a fresh thread when you're changing topics.
When you reply to an agent from iMessage or Slack, grasscoding routes that reply to the right thread automatically — you don't need to think about it.
Coding agents
Agents are the CLI tools that actually edit code. grasscoding supports the agents that consistently rank #1 on SWE-bench: Claude Code, Codex, Gemini CLI, and Cursor CLI. They all run inside the same computer, see the same filesystem, and follow the same project prompt configuration — so you can switch between them mid-project.
You authenticate each agent with your own provider account. grasscoding never resells model API access; you bring your own subscription (Claude Pro, ChatGPT Plus, Gemini Advanced, Cursor Pro) and pay providers directly for usage.