A teammate opens Claude Desktop, sees a new Cowork button, and asks the obvious question: is this just normal Claude Chat, or is it the same thing developers use in Claude Code? The confusion is natural. The same product family now contains several entry points with different permissions and different access models.
The difference matters. If you try to configure an API key inside Cowork where that setting does not exist, or if you try to process local files through a plain chat that cannot access them, the workflow breaks before it starts.
BetterToken works with AI coding tools and API access every day. For us, the useful first step is to separate three surfaces clearly: Claude Chat, Claude Cowork, and Claude Code.
What each mode is
Claude Chat is the ordinary conversational interface. It is good for questions, writing, analysis, summarization, brainstorming, and explaining concepts. It does not directly operate on a local project folder like a coding agent.
Claude Cowork is an agent mode inside Claude Desktop. The user points it at a folder or workspace and Claude works inside that boundary: reading files, editing documents, extracting information, updating tables, and handling multi-step tasks. The architecture is closer to Claude Code, but the interface is not a terminal. You can watch progress and intervene as the task runs.
Claude Code is the developer-focused agent. It works from the terminal, IDE, desktop app, and related surfaces. It reads codebases, edits files, runs commands, executes tests, and supports API Key plus Base URL configuration for third-party providers.
Comparison
| Capability | Claude Chat | Claude Cowork | Claude Code |
|---|---|---|---|
| Main use | Conversation, text, analysis | File and task automation | Software engineering |
| Local files | Uploads or selected context | Works in selected folders | Works in project directories |
| Command execution | No | Limited by product surface | Yes, with permissions |
| API Key / Base URL provider setup | No | Not the main path | Yes |
| Best audience | General users | Knowledge workers | Developers and technical operators |
| Typical output | Answer, draft, summary | Updated files or completed task | Code changes, tests, commits |
When to use Claude Chat
Use Chat when the work is mostly thinking or writing:
- summarize a document you paste or upload;
- compare options;
- draft an email, brief, or explanation;
- ask a conceptual question;
- brainstorm a plan before touching files.
Chat is the lowest-friction surface. It is also the safest when you do not want an agent to take actions on your machine.
When to use Claude Cowork
Use Cowork when the task is file-based but not primarily software engineering:
- clean up a folder of documents;
- extract data from PDFs into a table;
- update a report using several source files;
- reorganize meeting notes;
- prepare a draft from a set of local materials;
- run a multi-step office workflow while you supervise.
Cowork is useful because it brings agent behavior to non-developer work. It can operate across files without forcing the user into a terminal.
When to use Claude Code
Use Claude Code when the work involves a codebase, tests, shell commands, dependencies, or deployment:
- understand a repository;
- fix a bug;
- write or update tests;
- refactor a component;
- add an API endpoint;
- update documentation from code;
- run a linter or build command;
- work through a pull request.
Claude Code is also the surface where API provider configuration matters most. If you use an Anthropic-compatible gateway such as BetterToken, the coding agent needs the right Base URL, token, and model configuration.
API access and providers
This is the most common source of confusion. Claude Chat and Cowork are product surfaces tied to Claude app access. Claude Code can be configured for developer workflows and third-party provider routing. That is why setup guides for API Key and Base URL usually target Claude Code, not Cowork.
If your goal is to run coding tasks from Russia through an API gateway, you are probably looking for Claude Code setup. If your goal is to automate local documents without touching a terminal, Cowork may be the better entry point, but it is not a replacement for provider configuration.
Security model
Each mode should be treated according to the access it receives:
- Chat sees what you type or upload.
- Cowork may see and modify selected folders or connected resources.
- Claude Code may read and edit repository files and request shell commands.
For sensitive work, keep scopes narrow. Give the agent only the folder it needs. For code, start with manual permission and inspect diffs before accepting changes.
Which one should you choose
Choose Chat if the task is mostly language.
Choose Cowork if the task is office-style automation across files.
Choose Claude Code if the task is engineering work, involves commands, or needs API provider setup.
The naming is similar, but the operating model is different. Once you separate these modes by access and task type, the product becomes easier to explain: Chat thinks with you, Cowork works through files with you, and Code changes software with you.