Launched this week
SuperTurtle
Remote control for Codex, handled from your phone
70 followers
Remote control for Codex, handled from your phone
70 followers
Anthropic built Remote Control so you can manage Claude Code from your phone. We built the same thing for Codex. Message SuperTurtle on Telegram, by text or voice, and it runs Codex in a cloud sandbox. No laptop needed, no terminal open, just your ChatGPT subscription.
Congrats on the launch! 🚀
As a new user, what would be the easiest way to get started and see the main value of the product?
SuperTurtle
@julia_zakharova2
Hey Julia, thank you for the comment!
Just sign up at superturtle.dev, add your ChatGPT account, and you're ready to go.
We tried to really polish the onboarding, so it only takes about two minutes. Enjooy!
SuperTurtle
Congrats on the launch! This is very close to something I’ve been building for personal use. Curious about the TTS part: is that native to Codex, or did you build your own layer on top?
Also, where do you see the interface evolving from here? Could this work across multiple groups, with the bot helping support different projects in each?
@vladimir_mynka Hi Vladimir! We currently use OpenAI TTS, it's very cheap and works well. It's supported in the open source version since I haven't figured out yet how to safely store the OpenAI API key on the VM without the agent being able to read it.
Could you elaborate on the multiple groups set up? What would be the use case? Someone has already opened a PR to support Telegram groups, I'm just not sure yet how to do it technically in a clean way. :)
@rigos0 Got it, thanks. I’ve been thinking about this secret-storage problem a lot too. As far as I’ve been able to tell, this direction is usually described with terms like credential proxy and secretless or identity-based access — where the agent can use a credential through some intermediary layer without being able to inspect the raw value itself. I’ve seen a few off-the-shelf attempts in this direction on PH and GitHub, but most of them still seem fairly early-stage.
As for the groups setup: right now I just create multiple Telegram chats, add my bot to each of them, and keep a one-to-one mapping between chat_id and repository_url, so each chat is dedicated to a single repo and its Codex workflow. In my case, this is built around a local Codex app-server running on my laptop, with an extra custom orchestration layer on top. Conceptually, it feels like a cloud version could work in a similar way, although I’m less sure what the clean implementation would look like there.
Another direction, as I understand it, is the one proposed in the PR: using a single Telegram supergroup and assigning each project to its own forum topic, with routing based on message_thread_id
Hi everyone!
Really excited to launch SuperTurtle here! I am in charge of our infra. We did our best to make the onboarding extremely simple and to cover all VM costs for the first 100 people.
The telegram bot is very smart as it’s powered by the SOTA models from OpenAI.
Would glad to hear what works and what doesn’t!
Snack Prompt