繁體中文

← Back

Chapter 1 (free): why Claude Code belongs in your Telegram, and how the bridge works

The itch

Claude Code is the most capable assistant you own, and it lives in a terminal on one machine. Your questions show up on your phone, on the bus, at 11pm. Every "I'll do it when I'm at the desk" is a task that never happens.

Telegram is already the app you have open. A bot there is free, instant, and works on every device you own.

So: a tiny bridge. Message the bot → it runs claude -p "<your message>" in a folder on your computer → the answer comes back. Follow-ups continue the same Claude Code session, so "now make it shorter" works. Photos and files land in an inbox/ folder and Claude reads them.

Why not the API, a framework, or a hosted service?

The safety model in one paragraph

Only your Telegram user id gets answers. Claude Code runs in dontAsk mode with an allowlist of tools — read, search, edit files in your workspace, web search, a few read-only git commands. Anything else is auto-denied and Claude tells you what it wanted to do. You widen the list when you trust a workflow. The working directory can't leave the roots you allow. This is the difference between "an assistant" and "a remote shell for whoever guesses your bot name".

The first three steps (you can do these now)

What the paid pack adds

The bridge itself (bot.mjs), the one-click setup.sh with a self-test, background service install for macOS and Linux, the assistant CLAUDE.md template, per-chat sessions with /new /cwd /model /status /cancel, photo/file/voice inbox, and MISTAKES.md — 14 real mistakes from shipping a paid Telegram product with Claude Code.

Get the pack — $12