Quick share of a feature we built into Telechat (self-hosted Claude AI bot) that's been surprisingly useful for devops workflows: Desktop Bridge with mobile approvals.
You're running Claude Code (or any Claude-powered agent) on your workstation. It's refactoring a module, running tests, deploying to staging. You step away for coffee, a meeting, or just to stretch.
Claude hits a tool call that needs human approval:
rm -rf build/
(wants to clean the build directory)git push --force
(rebase gone wrong)kubectl delete pod
(scaling decision)Without you at the keyboard, it just... waits. For however long you're gone.
Telechat's Desktop Bridge connects your Claude Code session to your phone via Telegram, WhatsApp, or Slack. When Claude needs approval:
All from your phone. No VPN, no SSH, no laptop.
Unattended CI/CD with a human gate. Run Claude as part of your pipeline for code review, test generation, or deployment prep. Gate the destructive steps on mobile approval instead of blocking the pipeline until someone checks Slack.
Overnight tasks. Kick off a large refactoring or migration analysis before bed. If Claude needs a decision at 2 AM, you'll see it in the morning and approve from your phone. It doesn't lose context while waiting.
Pair programming while mobile. Reviewing Claude's work from your phone between meetings. Approve the good stuff, deny the risky stuff, add context via chat.
Telechat runs on your workstation alongside Claude Code. It acts as a bridge between Claude's approval prompts and your messaging app. When Claude's tool-use loop hits a human-approval checkpoint, Telechat intercepts it, formats the request, and sends it to your Telegram/WhatsApp/Slack. Your response flows back and unblocks the agent.
No cloud relay — the bridge runs locally.
npm install -g telechatai
telechat init # interactive setup
telechat bridge --connect # link to Claude Code
telechat.fyi | npm install -g telechatai
Curious if anyone else has solved the "approve AI actions remotely" problem differently.