Telegram Remote — Setup & Usage

If you don't have it yet, pick one:

macOS & Linux:

curl -sSf https://contextguard.tech/install.sh | sh

Windows (PowerShell):

irm https://contextguard.tech/install.ps1 | iex

More options on the CLI installation page.

Easiest way — from the dashboard: open Settings → Telegram Remote Control and click "Connect Telegram". It opens the bot — press "Start" and your account is paired.

This only links your account to the chat. The local bridge itself is started separately — step 3.

Grab a key from the API Keys page and pass it to either of these once:

contextguard tg --api-key cg_live_...

or

contextguard remote --api-key cg_live_...

The CLI saves the key to ~/.claude/contextguard/remote-auth.json on your machine. After that you never need the key again — just run contextguard remote with no flags, in any new terminal window, any day. You don't need a new key every time you open a fresh terminal — the old $CONTEXTGUARD_API_KEY environment variable only lives inside one window; the saved file doesn't.

Option A — from a terminal:

contextguard remote

Option B — inside Claude Code itself, no separate window:

/plugin marketplace add ChevvyOkK/contextguard-plugin /plugin install contextguard@contextguard /contextguard:remote

The first two lines are one-time (install the plugin). /contextguard:remote starts the bridge in the background of your current Claude Code session and tells you whether it connected.

You're ready when you see: Connected. Waiting for prompts from Telegram...

While the bridge is running, just message the bot what you need — like any other chat:

fix the failing test in auth.service.spec.ts

You'll get a live-updating message: first status lines like "🔧 Tool: Edit: src/auth.service.ts", the answer text growing in as it's generated, and at the end the actual answer plus a recap of what was read/changed.

Useful chat commands: /status — active sessions, /stop — interrupt the current task.

Prompts arriving from Telegram run without interactive prompts — there's no one on the other end to answer "y/n". So file reads and edits (Read/Edit/Write) happen automatically, and anything not explicitly allowed (arbitrary Bash, for instance) is silently denied by default. This only applies to sessions started via Remote — your normal interactive Claude Code terminal session is unaffected.