[ BETA · Pro ]
Claude in Vaulterm Code,
on your SSH server.
A Claude panel in place of a terminal tab, built on the Claude Agent SDK. It is in the Beta channel (version 0.2.0-beta.1) and needs Vaulterm Pro.
Illustration
I will run the tests, then fix the failing one.
pytest -qAsks before it runssrc/app.pyDiff, Accept, RevertThis message contains a secret from your vault.
Remove, or Send anywayHow it works
- Replies stream in as Claude writes them.
- Every command and every edit shows as a card, so you can see what Claude did.
- Claude asks before it acts.
- Mention files with @, and pick Sonnet, Opus or Haiku.
- Resume past sessions.
- Slash commands, CLAUDE.md and the project's .claude commands and skills work, through the Claude Agent SDK.
On your server
On a project on a server, Claude Code runs on the host, in the project folder, over the SSH exec channel. Your API key goes over the channel's stdin, so it never appears in ps, in a file or in a log.
If Claude Code is missing on the host, Vaulterm offers to install it and asks first. It also asks the first time for each host and folder before Claude runs commands there.
Undo anything
Each reply lists the files it changed, with a diff, and Accept and Revert for each. You can also rewind to any earlier message and the files come back as they were. Checkpoints are snapshots that do not touch your repository.
Privacy gate
A message that contains a secret from your vault is stopped before it is sent. You can Remove the secret or Send anyway.
Reading .env files, SSH keys, credentials and key or certificate files always asks first, even in Bypass mode and even if you saved an Always allow rule. The vault itself is never put in Claude's context.
What does leave your computer: your prompts, the files you mention and tool results go to Anthropic under your own key. Vaulterm receives no copy. Anthropic's terms for your API key apply: Anthropic's commercial terms.
Permission modes
The default. Claude asks before a command or an edit.
File edits go through. Commands still ask.
Claude plans and does not change anything.
Claude runs commands without asking. Vaulterm asks you to confirm before it turns on.
MCP servers
Add your own MCP servers, stdio, http or sse, up to 40, in Settings, Code, Claude. A server never holds a secret: its environment names a vault Environment, and Vaulterm reads the values at send time. How Environments work.
What you need
- Vaulterm Pro. Claude is Pro for local folders and servers alike.
- The Beta channel: Settings, Updates, Channel.
- Your own Anthropic API key, stored in your vault. Anthropic bills the usage to you.
- Claude Code installed where the code is. Vaulterm finds it; on this computer it shows the install command to copy, and on a server it offers to install it after asking.
There is no Claude login or subscription sign-in, and no Bedrock or Vertex.
Questions
Is my code used for training?
Anthropic's terms for your API key apply. They are at Anthropic's commercial terms.
Does it work on local folders?
Yes, on Pro. Claude is a Pro feature for local and server projects alike.
How is this different from Agents?
Agents runs the Claude Code, Codex or Gemini CLI in a terminal tab and watches them. Claude in Vaulterm Code is its own panel, with cards for each command and edit. See Agents.
Claude and Claude Code are trademarks of Anthropic. Vaulterm is not affiliated with or endorsed by Anthropic. More on setting it up: the Beta channel.