Why does this matter?

There is no single "Claude". You have the regular chat on claude.ai, the agentic Cowork mode, and Claude Code for actually building, and the last two each exist in a web and a local variant. People lump them together and then wonder why one can "reach their files" and the other can't. Keep the forms apart and you pick the right one faster, and you know exactly what happens with your files and data.

These forms evolve fast and some are still young. The underlying logic is stable, though, and that is what we put front and centre here.

The two questions that decide everything

With those two questions, every form falls into place.

The forms at a glance

Everything side by side

ModeWhere it runsAccess to your local filesWhat for
claude.ai Chat / Project Vendor's cloud No, only what you type or upload (instructions + context) Quick conversation, writing, no-code
Cowork, web/mobile Cloud session (keeps running in the background) No, only indirectly via the desktop app as a bridge Agentic knowledge work, building something without code
Cowork, desktop app Session in the cloud, the app is the bridge to your device Limited: only connected folders, and only while the app is open. Plus local connectors/MCP Same as Cowork web, with access to local files and tools
Claude Code, local (CLI / VS Code / desktop extension) Your machine (local) Yes, full & native (the folder/repo you open) + commands, git, tools, MCP Real projects: files, systems, automation (incl. software)
Claude Code, web/cloud Cloud sandbox No, works on a cloned GitHub repo in the cloud Same as local, but without your own machine

Web or local: where's the real difference?

"Web versus local" means something different for Cowork than for Claude Code.

In short: with Cowork, web-versus-app is about whether there's a bridge to your device; with Claude Code, it's the real difference between your files and a copy in the cloud.

When do you choose what?

Two questions settle it: where does it run (your machine or the cloud) and what is it for (knowledge work or real projects). Everything else follows from that.

How Sevendays uses this: for knowledge work and quick tasks we stay in chat, projects and Cowork; for real software, integrations and automation we use Claude Code locally on the real repo. That's how we build and run our AI platform and the things you see in our examples. Want to know how you run the same model as a subscription or an API key, and where the application around it runs? That's covered in Claude subscription or API key?. Not sure what fits you? Our approach to AI always starts from your concrete situation.

Related terms