Stop babysitting your AI coding agent. Let it text you.
Agentic coding has a dirty little secret: you spend half the time watching the agent work. You give it a task. It plans, edits five files, runs the tests... then stops to ask "PostgreSQL or SQLite?" You're in the kitchen. Ten minutes gone, for a one-word answer. The fix is simple: let the agent text you. What that looks like You give Cortex AI IDE a task and walk away. Your phone buzzes on Telegram: Cortex needs your answer Which database should the app use? 1. PostgreSQL 2. SQLite You tap 2. The run carries on. A few minutes later: ✓ Done in 41s · 12 tests passed It works the other way too. Send a task from your phone as a text, a screenshot of a bug, or a voice note on the way home, and it runs on your computer. "Wait, isn't that dangerous?" Fair question. Here's how it's locked down: The phone can't approve commands. It can answer questions, but anything that actually runs still waits for you at the computer. No open ports. No public
Agentic coding has a dirty little secret: you spend half the time watching the agent work.
You give it a task. It plans, edits five files, runs the tests... then stops to ask "PostgreSQL or SQLite?"
You're in the kitchen. Ten minutes gone, for a one-word answer.
The fix is simple: let the agent text you.
What that looks like
You give Cortex AI IDE a task and walk away. Your phone buzzes on Telegram:
Cortex needs your answer
Which database should the app use?
1. PostgreSQL
2. SQLite
You tap 2. The run carries on. A few minutes later:
✓ Done in 41s · 12 tests passed
It works the other way too. Send a task from your phone as a text, a screenshot of a bug, or a voice note on the way home, and it runs on your computer.
"Wait, isn't that dangerous?"
Fair question. Here's how it's locked down:
- The phone can't approve commands. It can answer questions, but anything that actually runs still waits for you at the computer.
- No open ports. No public IP, no webhook, nothing exposed.
- Only your chat works. Messages from anyone else are silently ignored.
The other half: any model, even a free local one
The agent doesn't need an expensive API. With Cortex Connect, you can point it at anything OpenAI-compatible: Groq, OpenRouter, or Ollama running on your own laptop. Your keys never leave your machine.
A local model doing the work, and your phone in your pocket. The code keeps moving while you get on with your day.
Try it
- Setup takes about 5 minutes: Telegram guide (the bot is part of the Pro plan; the IDE itself is free)
- Download: cortex-ide.app
How do you handle long agent runs? Still babysitting, or have you found a better way? ????
Disclosure: I'm a developer of Cortex AI IDE at Pasons Group.