Where Your Agent Can Live
Claude comes in several forms and they are not equivalent. This page exists because picking the wrong one costs you most of what makes the system worth using.
The short version
Section titled “The short version”Use VS Code. It is free, it is the most documented, and it gives the system its full capability. Cursor and Antigravity are equally good if you already prefer one of them.
The tiers, honestly
Section titled “The tiers, honestly”Full capability — VS Code, Cursor, or Antigravity
Section titled “Full capability — VS Code, Cursor, or Antigravity”These are code editors. They are the interface you use to talk to your agent, and they are the only option that gives you the complete system.
You get persistent memory in files on your machine, full customisation of how the agent behaves, the specific memory banks and training the system builds for you, sub-agents, and external tool connections.
This is what the setup guides are written for. All three work identically for our purposes — the differences between them do not matter here.
Partial — Claude Cowork, in the Claude desktop app
Section titled “Partial — Claude Cowork, in the Claude desktop app”Cowork can install skills. What Claude has built there is the user interface, not the full agent environment.
You can point Cowork at the GitHub repository and get real use out of it. What you do not get is the same depth of customisation, memory, and training that the editors provide. There are no dedicated Cowork plugins for the 90MM BOS today, and there may never need to be — the repository itself is the install.
Least — Claude Projects, on claude.ai
Section titled “Least — Claude Projects, on claude.ai”Projects gives you the least of the three tiers. It can hold documents and context, but it cannot provide the file-level memory and customisation the system is built around.
If Projects is genuinely all you have access to, you will get something from reading the methodology. You will not get the operating system.
Why the editors win
Section titled “Why the editors win”The system’s memory lives in files, not in chat history. Your constraint, your personality profile, your decision log, your context — all of it sits in a folder that the agent reads at the start of every session.
That is why a new conversation does not start from zero, and it is why the editor tier matters. An interface that cannot read and write that folder cannot run the system, however good the model behind it is.
Is this the same Claude I already pay for?
Section titled “Is this the same Claude I already pay for?”Yes. A Claude Pro or Max subscription powers Claude Code — no second account, no second subscription.
But the interface is genuinely different:
Claude.ai is a chat. You ask, it answers, and each conversation starts fresh with no memory of the last one.
Claude Code reads your business files, remembers your constraint, knows your communication style, and gets sharper over time.
Texting a smart friend, versus that friend moving into your office and reading everything.