Claudex — Claude Code and Codex AI chat for VS Code
Two coding agents. One place to work.
Use Claude Code and Codex beside your code, with your existing CLI accounts and settings. Ask questions, turn a plan into working code, and review changes in VS Code. Continue a conversation with another agent and choose what context to carry forward.
Why Claudex?
- Keep the context when you switch agents. Continue a thread with Claude Code or Codex and inspect the context chip before sending.
- Go from plan to implementation. Review and edit a Markdown plan, then start implementation from the latest version.
- Review the work as it happens. Respond to provider approval requests and open changed files in VS Code’s diff editor.
- Give each task its own space. Keep local conversation threads and optionally start a task in a Git worktree.
- Bring your existing setup. Use your installed agents’ skills and MCP configuration, with separate profiles for additional accounts.
Requires Claude Code or Codex installed and signed in. Provider usage limits still apply. Developed and tested on macOS; other platforms have not yet been validated.
Get started
Requires VS Code 1.96+, a trusted local project folder, and Codex or Claude Code installed and signed in. Usage follows your existing provider account and limits.
- Install Claudex from Extensions, or run Extensions: Install from VSIX… to install a downloaded VSIX.
- Open your project folder and run Claudex: Open Workspace from the Command Palette. You can also click the sparkle button in the editor title bar or status bar.
- Chat opens beside your code. For the larger layout shown below, run Claudex: Open in Editor.
If an agent isn't found, set agentWorkspace.codexPath or agentWorkspace.claudePath to its executable's absolute path.
How to use it
Screenshots show the current UI with sample conversations.
1. Start a chat
Click + for a new thread. Choose your agent and model from the model menu below the message box; Default uses your CLI's configuration. Pick a mode:
| Mode | Use it to |
|---|
| Ask | Ask questions about your code. Edit tools are disabled. |
| Plan | Research a task and prepare a plan before building. |
| Auto | Work on a task with the CLI's approval reviewer. It asks you when it cannot decide. |
Claude's Ask mode can still run shell commands allowed by your Claude settings. New chats start in Auto. If the provider cannot apply it, the mode label shows what is actually running. Auto does not guarantee safety; state any limits in your prompt.
Type your request and press Enter. Use Shift+Enter for a new line. Type @ to add files, folders, threads, or other context; use the paperclip to attach images. Type / to choose a skill.
Use multiple Claude and Codex accounts
Keep personal and work accounts in separate profiles. Open the model selector → Manage account → Add account, choose Claude Code or Codex, name the account, and keep Create a separate account selected. Then use Open sign-in terminal to complete the CLI’s own sign-in.
After signing in, choose the account in the model selector, then choose its model. Your unsent draft stays in place.
Still images and account instructions
2. Review a plan, then build
Select Plan and describe the task. Review the plan inline, use Open plan to read it in VS Code, or Edit plan to edit its Markdown. Reply in chat to request changes or add notes in the composer. Click Implement in Auto when ready; it reads your latest edits and switches the thread to Auto.
3. Review requests and changes
When a request appears, read the command or action, then choose Allow once or Decline. Open Changes in the header to see uncommitted files in the thread's checkout. Click a file to open VS Code's diff editor.
Use the tabs or Chats to return to a thread. Closing a tab keeps the conversation and any running work. To continue with another agent, choose it from the model menu, inspect the Context from … chip, and press Enter to send.
More visual guides
Choose a model and reasoning effort — Instructions and static image
These short clips use sample conversations. Native editor actions stop at the webview boundary; no provider work runs during capture.
Read rich answers in Ask mode — Instructions and static image
Understand diagrams and visual explanations — Instructions and static image
Answer agent questions — Instructions and static image
Review approval requests — Instructions and static image
Keep multiple chats and find past work — Instructions and static image
This is the beginning of the README. Read the rest at the original. Read the full README ↗
