Use Claude in Vaulterm Code (beta)

The Claude panel sits in Vaulterm Code’s side bar. Replies stream in, every command and edit shows as a card, and Claude asks before it acts. It works on a folder on this computer or on a saved server. For what it is, see Claude in Vaulterm Code.

Switch to the Beta channel

Open Settings → Updates. Before 0.2 this section is called About.
Under Channel, choose Beta.
Install the update when Vaulterm offers it, and restart.

You can go back to Stable in the same place.

Add your Anthropic API key

Open Settings → Code and find Claude.
Press Add key, paste your key into Anthropic API key, and press Save key. To make a key, use Create a key at console.anthropic.com.

The line above the button reads “API key: set” once it’s saved. The key is stored in your vault and is never shown again. Change replaces it and Remove deletes it from the vault. The key at Anthropic stays valid.

Claude needs Pro, on folders on this computer too. Without Pro, the section says “Claude built into Vaulterm Code is part of Vaulterm Pro”, with Unlock with Pro.

Claude uses only this key. It never uses a Claude login on your computer.

You can also add the key from the panel itself: with none set, it asks for one.

Install Claude Code

Vaulterm runs your own copy of Claude Code. It looks for claude on your path and in the places Claude Code’s installers use.

On this computer, if it isn’t found, the panel says “Claude needs Claude Code installed on this computer” and shows Anthropic’s install command. Press Copy command, run it in a terminal, then press Check again.
On a server, the panel says “Claude Code isn’t installed on” the server. Install on this server installs it there with Anthropic’s installer. It asks first, and nothing is installed without your yes. Then press Check again if needed.

The first time Claude runs on a server and folder, the panel says “Claude will run commands on” the server. Press Allow Claude on the server to go on, or leave it off. Your API key is sent to the server over your SSH connection and lives in the Claude process’s environment there while Claude runs.

Open the Claude panel

Press Claude, the sparkle icon in Code’s side bar. New session starts a conversation, and Sessions lists past ones, each with Resume, Rename, Fork and Delete.

Mention files

Type @ in the message box to mention a file. The list also offers @selection (the editor’s selection), @terminal (the terminal’s selection or last lines), @changes (the working tree changes) and @problems (the Problems list). Type / at the start for the slash commands Claude Code reports. Press Enter to send and Shift+Enter for a new line.

Pick a model

Choose Model under the message box: Sonnet, Opus or Haiku. Each project remembers its choice. Settings → Code → Claude has Default model for new projects. You can’t change the model while Claude is working.

Permission modes

Choose Mode under the message box:

Mode	What Claude may do without asking
Ask	Nothing that changes your project. It asks first. This is the default.
Accept edits	Edit files. It still asks before running commands.
Plan	Plan only, without making changes.
Bypass	Run any command without asking. Vaulterm asks you to confirm, and shows “Bypass is on” while it’s active.

When Claude asks, its card has Allow once, Always allow for this project and Deny. Always allow saves a rule in .vaulterm/claude.json in the project. When you deny, you can type a reason, and Claude sees it.

With a question waiting, Enter allows it once and Escape denies it. A new mode applies from your next message.

Accept, Revert and Rewind

After each reply, Changes in this turn lists the files Claude changed, with a diff for each.

Accept keeps a file, and Accept all keeps them all.
Revert puts a file back as it was before the message, and Revert all does it for every file. Both ask first.
Rewind to here, on an earlier message, puts your files back as they were then, and continues the conversation from before that message. It asks first.

The privacy gate

A message that contains a secret from your vault is stopped before it is sent. Remove the secret and send again, or press Send with the secret redacted.

When output from a tool contains something that looks like a secret, it is shown redacted, with Reveal (which asks first).

Reading a .env file, an SSH private key, a credentials file, or a key or certificate file always asks, even in Bypass, and an Always allow rule doesn’t skip it. Claude has no tool that reads your vault.

Your prompts, the files you mention and tool results go to Anthropic, under your key. Anthropic’s terms for your API key apply. See Security.

MCP servers

MCP servers give Claude more tools. They are managed in Settings → Code → Claude, under Claude: MCP servers, and they run on this computer only: on a server, the panel says “MCP servers are off on servers”.

Press Add MCP server.
Give it a Name: letters, numbers, - and _, up to 40 characters.
Choose its Type: stdio, http or sse.
For stdio, enter the Command that starts it. For http and sse, enter its URL, which must start with http:// or https://. Keep secrets out of URLs.
If it needs a secret, choose an Environment from your vault. For stdio, its variables become the server’s environment variables. For http and sse, they become request headers, with the variable’s name as the header’s name.
Choose whether it is for All projects or one project, and press Add server.

You can add up to 40 servers. A server starts when you send a message, and its status shows in the panel’s header. Tick the box beside a server to switch it on or off, and use Edit or Remove to change it.

If something doesn’t work

“Claude needs Claude Code installed”: install it as above, then press Check again. If it is installed somewhere unusual, make sure claude is on your path.
No key: the panel asks for your Anthropic API key. Add it as above.
A message that says Claude comes with Pro (NeedsPro): your license doesn’t include Pro, or it ended. See Pricing.
A message stopped: it holds a secret from your vault. See The privacy gate.