Everyone can control Claude Code from a phone now. Then what?
The remote-control layer of agentic coding got commoditized in one quarter. The layer above it — deciding what the agent works on — is where the leverage moved.
A solved problem
If you want to drive Claude Code from your phone in 2026, you are spoiled for choice. Anthropic shipped Remote Control (steer your local session from the mobile app) and then Channels — official Telegram, Discord, and iMessage bridges into a running session, permission prompts included. Around them: Omnara's polished mobile command center, Happy Coder's free end-to-end-encrypted client, and a shelf of open-source Telegram bots. All of them work. Most of them are free.
That's what commoditization looks like: in about one quarter, “message your coding agent from your phone” went from a product to a feature to a platform default.
The question none of them answer
Every one of those tools starts after the interesting decision. You still notice the problem, decide it matters, scope it, and type it. The agent has infinite patience and 24/7 availability; the input stream is still one founder's attention. Steering got easy — sourcing didn't.
And the highest-signal source of worth-doing tasks isn't your backlog. It's the customer who just hit the bug, on your site, right now, telling you about it in their own words.
The layer above: from complaint to PR
That's the layer chatwithdev occupies. The chat widget on your site relays every conversation to your own Telegram. When a report is real, you tap ➕ Ticket — the ticket is drafted from the transcript and confirmed by you, so the agent only ever receives your words. One more tap dispatches it: to a small runner on your VPS, or to GitHub where the Claude GitHub App opens the PR with zero infrastructure. You review and merge from your phone; the widget can tell the customer it's fixed.
The layers compose
This isn't either/or. Keep Channels or Omnara or Happy for interactive work — pair-programming a feature from the sofa. Let chatwithdev feed the unattended work: the queue of real customer-reported issues that your box grinds through one-shot, on branches, while you're doing something else. Same VPS, same Telegram app, two different doors. The honest comparisons are here: vs Claude Code Channels, vs Omnara, vs Happy Coder.
Getting started
The foundation is the chat relay, and it's live today: start free, paste one script tag, and website chats arrive in your Telegram. Tickets and dispatch are rolling out now — the VPS guide walks the whole loop.
FAQ
What's the difference between chatwithdev and Claude Code Channels?
Channels pushes messages from your own Telegram into a Claude Code session you have running — it's how you talk to your agent. chatwithdev is the customer door: visitors chat on your website, reports become durable tickets, and you dispatch them to your agent with a PR as the gate. They compose; they don't compete.
Why not let customers talk to the coding agent directly?
Because customer text should be treated as data, not instructions. In chatwithdev the dispatched task is always the founder's confirmed ticket; the customer transcript is attached and explicitly marked as untrusted context. That's a security boundary, not a limitation.
Which dispatch tier should I start with?
If your repo is on GitHub, start with GitHub dispatch — a ticket becomes an issue mentioning @claude and the Claude GitHub App opens the PR, no infrastructure needed. Move to the self-hosted runner when you want jobs on your own VPS with your own agent command.