vibekit-agent

Control Claude Code or Codex on your own computer from the VibeKit iPhone app. Ship code from your phone.

This is VibeKit's local bridge mode. Your code runs on your machine, not in VibeKit cloud.

Free to use - works with your existing Claude Pro/Max subscription or your ChatGPT plan.

Installation

npm install -g vibekit-agent

Quick Start

1. Link your computer:

   npx vibekit-agent link

   In the VibeKit app tap the + beside the dock (or Add new workspace at the end of Home in older versions), then Connect a computer (in app versions before 1.5, go to Remote and tap Connect a computer). A paired computer shows on Home and counts as one of your plan's workspaces. The app walks the same steps as the terminal: when the terminal prints a QR code, tap Open camera and scan it (your iPhone camera works too); when it asks for a code, the app shows one to type. Then press Enter to start the agent right away. Can't scan? Tap Type the code instead and enter the code printed under the QR. For Codex, run npx vibekit-agent link codex.

2. Older VibeKit app? Tap Pair an agent in Remote and type the code it shows into the same terminal prompt.

3. Starting it later:

   npx vibekit-agent start

4. Send messages from the Remote tab - Claude Code or Codex runs them on your machine.

Claude Code or Codex

Name the coding agent when you link: npx vibekit-agent link claude or npx vibekit-agent link codex. Without one, link picks Claude Code if it is installed, otherwise Codex. To switch later (remembered for next time):

npx vibekit-agent start codex
npx vibekit-agent start claude

Codex runs on your own ChatGPT account. If it isn't installed or signed in, link codex and start codex offer to do both at the terminal. To do it yourself:

npm install -g @openai/codex
codex login

When to Use This

Use vibekit-agent when you want:

• to control Claude Code or Codex on your own laptop or desktop from your iPhone
• to use your local Claude or ChatGPT sign-in instead of VibeKit cloud execution
• a remote-control bridge for local development

Do not use it if you just want VibeKit-hosted tasks or deployments. For that, use vibekit-cli or vibekit-mcp.

Features

• Text messages - Chat naturally, the agent responds
• Voice messages - Send voice notes
• Images & screenshots - Send photos for analysis or debugging
• Document uploads - Share files for the agent to work with
• Conversation memory - The agent remembers context within a session
• Ask before actions - Approve or deny each command and edit from the app or the lock screen
• Continue a desk session - Pick up a Claude Code session you started in the terminal on this computer from the app (1.8.0+). It is a handoff, not a mirror: reopen it at the desk with claude --resume to see what you did from your phone

Commands

Command	Description
vibekit-agent link	Link this computer to your VibeKit account
vibekit-agent start	Start the remote agent
vibekit-agent start -d /path/to/project	Start in a specific directory
vibekit-agent link codex	Link this computer and run Codex on it (or link claude)
vibekit-agent start codex	Switch this computer to Codex (or start claude)
vibekit-agent status	Show connection status
vibekit-agent logout	Remove stored credentials
vibekit-agent config	View current configuration

Tool Permissions

By default the agent has full access: it edits files and runs commands on your behalf, because you linked the computer for that. You can restrict it:

# View current tools
vibekit-agent config

# Set specific tools
vibekit-agent config --tools "WebSearch,Read,Glob,Grep"

# Reset to a read-only set
vibekit-agent config --reset-tools

# Allow all tools again
vibekit-agent config --all-tools

Available tools: WebSearch, WebFetch, Bash, Read, Write, Edit, Glob, Grep

Codex has no native per-tool list, so it honors two settings: full access (all tools), or read-only (a list with no Write, Edit or Bash). In read-only mode VibeKit removes Codex's shell tool and supplies Read, Grep, Glob and LS tools confined to the selected working directory. A list that allows edits but not commands, or the reverse, is refused.

For Ask before actions, Remote starts Codex with a clean policy profile that shares your login but does not load local “always allow” command rules. This keeps every write and non-read command on the phone approval path.

How It Works

1. The agent runs on your local machine
2. It connects to VibeKit's server via WebSocket
3. When you send a message from the iOS app, it's relayed to your local agent
4. Claude Code or Codex executes the request and sends the results back to the app

Requirements

• Node.js 18+
• Claude Code CLI (npm install -g @anthropic-ai/claude-code) or Codex CLI (npm install -g @openai/codex)
• A VibeKit account (free to start)

Troubleshooting

"No token found" - Run vibekit-agent link first

"Connection lost" - Check your internet connection and restart with vibekit-agent start

"Claude Code not found" - Install Claude Code: npm install -g @anthropic-ai/claude-code

"Codex isn't installed on this computer" - Install Codex: npm install -g @openai/codex. The agent skips a codex that fails codex --version, so reinstall a broken one.

"Codex isn't logged in on this computer" - Run codex login on that computer

Related Packages

• vibekit-cli: terminal client for VibeKit cloud workflows
• vibekit-mcp: MCP server for VibeKit cloud workflows

Links

• VibeKit Website
• GitHub

License

MIT