I kept losing my Claude Code conversation every time I left my desk.
So I built onthego. One command moves your uncommitted changes and your Claude Code or Codex session to your laptop. Resume there and it picks up mid-thought.
Free and open source.
https://t.co/drahO9DKsQ
i use A LOT of "chat mode", for learning and ideation,
and the claude chat mode is pure trash compared to GPT chat modes.
Why aren't people talking about this (for the last years) ?
I made a remote control mobile app for codex cli.
I had to use my VPS as a relay server. A process on my mac scans all codex cli sessions when turned on.
Tools like Grok, Muse, and Dots are all great, but fundamentally, I don't think relying on a specific vendor for a cloud-based personal agent is a good idea.
A personal agent needs to serve as a personal wiki and hold a metric ton of custom configurations. To really get the most value out of it, your daily life ends up relying heavily on it. Handing all of that over to a specific vendor’s shared cloud - where a single canceled subscription means losing everything overnight - feels like keeping your house deed in a subway locker.
Even if it's a huge pain, I believe it's far better to build your own environment from scratch and write code for your own customized setup. Sure, someone might argue, "Isn't a VPS provider just another vendor?" But with backups and snapshot features, that's more like a managed safety deposit box, not a sketchy subway locker.
I used to think all software would eventually be swallowed up as basic features of frontier AI labs, but recently I’ve realized the critical importance of third-party SaaS that remains independent of these AI providers.
Team is heading to Helsinki for the main event : we won Junction Korea.
During the event in Pohang, we were locked in at a motel near POSTECH, right in the middle of an area packed with bars and motels. We basically lived on jjajangmyeon, stayed up all night crammed into a single motel room, debating agentic systems and grinding away to push our benchmark scores higher.
I can confidently say it was a once in a lifetime experience.
I was wondering how I could utilize Astra more ..
I made a fork-style subagent delegation skill for Astra and Sol, via dynamically tuning "fork_turns". I think this "fork_turns" based v2 delegation is one of the most strong features of current Codex.
The universal Sol implementers are owned by Astra, and they consistently spawns in the thread when needed. The root Astra owns every usability touchpoint, from UI/UX and design to visual QA, since I think Astra is way more reliable than Sol for those kind of tasks ..
Thanks to @claudeebum, I no longer need all the weird workarounds, like running a background terminal for Pi. That said, I opened some PRs to let CURSOR BROS use Grok 4.5 Fast through OpenCodex. If they land, it’s going to be huge. Running Ralph inside Codex with Grok 4.5 Fast would be an absolute sight to behold.