@dmsdc-ai/aigentry-telepty

Cross-machine PTY-based remote prompt injection daemon for AI CLIs.

telepty (Tele-Prompt) is a lightweight background daemon that bridges the gap between the network and interactive AI command-line interfaces. It allows you to seamlessly share, attach to, and inject commands into terminal sessions across different machines.

Its primary user experience is prompt-driven operation inside LLM CLIs and the built-in TUI. Raw telepty ... commands are the lower-level control surface.

One-Click Installation & Update

To install or update telepty on any machine (macOS, Linux, or Windows), just run the command for your OS. (Node.js will be automatically installed if you don't have it).

For macOS and Linux (Ubuntu, CentOS, etc.)

Open your terminal and run:

curl -fsSL https://raw.githubusercontent.com/dmsdc-ai/aigentry-telepty/main/install.sh | bash

For Windows (PowerShell)

Open PowerShell as Administrator and run:

iwr -useb https://raw.githubusercontent.com/dmsdc-ai/aigentry-telepty/main/install.ps1 | iex

You can also launch the installer through npm without downloading the script first:

npx --yes @dmsdc-ai/aigentry-telepty@latest

These single commands will install the package globally and automatically configure it to run as a background service specific to your OS (systemd for Linux, launchd for macOS, or a detached background process for Windows). The installer now stops older local telepty daemons before starting the new one, so updates do not leave duplicate background processes behind.

Seamless Usage

1. Start a background session:

   telepty spawn --id "my-session" bash

2. Attach to a session (Local or Remote):

   telepty attach

   telepty will automatically discover active sessions on your local machine and across your Tailscale network!

3. Inject commands remotely:

   telepty inject my-session "echo 'Hello from nowhere!'"

4. Universal CLI submit (split_cr):

   All AI CLIs (Claude, Codex, Gemini) submit reliably via the split_cr strategy — text is injected first, then \r is sent separately after a 300ms delay. This works universally across all CLIs without any per-CLI workarounds.

   # The inject API handles split_cr automatically
   curl -X POST http://127.0.0.1:3848/api/sessions/my-session/inject \
     -H "Content-Type: application/json" \
     -d '{"prompt": "your command here"}'

CLI commands such as list, attach, inject, rename, multicast, and broadcast now auto-discover sessions across your Tailnet by default. If the same session ID exists on multiple hosts, disambiguate with session_id@host.

Testing

Run the full regression suite locally:

npm test

Keep the suite running while you work:

npm run test:watch

The automated suite covers config generation, daemon HTTP APIs, WebSocket attach/output flow, bus events, session deletion regressions, and CLI smoke tests against a real daemon process.

If the local daemon ever gets stuck or duplicated, open telepty and choose Repair local daemon.

Skill Installation

The package installer opens the telepty skill TUI automatically when you run it in a terminal.

To reopen it later, run telepty and choose Install telepty skills.

The TUI lets you choose:

• which packaged skills to install
• which target clients to install into (Claude Code, Codex, Gemini)
• whether each target uses a global path, the current project path, or a custom path