I kept losing track of my own agents. Six Claude Code and Codex sessions running, some on my main PC, some on the machine in the next room, and no clue which one was waiting on me.
So I built the workspace I wanted. It's open source now. It's called Jamat π§΅
spending this week adding Codex support to Jamat for OpenAI Build Week.
the whole thing was built around a pluggable agent layer, so Codex was always meant to slot in. it's running now, still polishing the edges.
Codex and Claude Code in the same workspace, finally.
I kept losing track of my own agents. Six Claude Code and Codex sessions running, some on my main PC, some on the machine in the next room, and no clue which one was waiting on me.
So I built the workspace I wanted. It's open source now. It's called Jamat π§΅
Honest status: Windows is first-class, macOS and Linux are beta. Solo spare-time project I've used daily for months, rough edges included.
MIT, self-hosted, your own keys, nothing proxied.
Link below π
Gpt 5.6 vs opus vs fable debate barely matters imo. The models are all good now.
What actually decides your day is the harness. and claude code's is light-years ahead of codex. Codex is just painful to live in.
@simonw honestly the product names are the easy part. the real maze is the model tiers. 5.6 vs pro vs sol/tera/luna, and somehow there's a Pro for Luna. it's names all the way down.
@ChrisLaubAI@DarioAmodei This is one of the many reasons I built Jamat :-). Claude is great, but there are a lot of small things that can be improved.
Feel free to check it out. Published it a few days ago, but I've been building it since Claude Code first dropped:
https://t.co/87TOpTkbu8
First post, so, quick why-this-exists.
I use Claude Code enough that I had multiple agents running and kept forgetting which ones were waiting on me.
So over the last few months I built a way to see them all at once. It's called Jamat, it's open source.