3code is becoming a stable workhorse
I couldn't be more excited- 3code's second beta release 0.7.0 is out!
After adding the sandbox and becoming feature complete, I made a bunch of improvements using just standard benchmarking tasks to be able to do it more quickly and when to solving 9 of 10.
In everyday use, 3code is just a joy- if you're a terminal guy like me, it's like vibecoding with vim (but it does support some emacs key combinations thanks to the first 3code contributor, Jory!
I am finding that using 3code is vibecoding in the most positive sense of the word. You know it's as efficient as can be, so all that inference you're running is being put to the best possible use. Its display is flickery only in one line, the token bar- and that reflects directly what the model is doing, so you can take its pulse, so to speak. I can tell which model I'm running from the token bar- kimi kind of speaks in blocks while GLM is all furious reasoning since the z.ai infrastructure updates.
And you get an interface that's just really simple- it's just a terminal. It's like running postgres. You get just two lines of interface that reall count, the token bar and the thinking ticker, and you can turn those off. Token usage gets logged with every turn so you know what the burn is.
The numbers speak for themselves.
Our latest release is 0.7.0 get it while it's hot! https://3code.capocasa.dev
And we've added comments! What do you think about 3code? let me know!