OK, so I don't know if you saw my previous comment from a while ago on the awesome-list - the first time i tried this, i liked it but it felt like really hard to control and things were running off the rails and i was a bit overwhelmed/confused. to be honest, i still kinda feel that way a bit lol - maybe it just takes some practice, but i tried it out again with the latest version on a small greenfield thing i'm playing around with, and it's really exciting to be honest - prob the most interesting thing i've tried in a long time. i had 3 agents going and they're communicating with each other, and doing a pretty good job of not getting lost or stepping on each other's toes, correcting course if they get mixed up (ok i'm using Opus so maybe this helps), and just got a ton of stuff done so quickly. i was hardly able to pay attention and it was a toy app so i kind of just let them go, but they seem to do an amazing job of coordinating things. (I think, at least, i haven't gone back to read the code yet.)
how's progress going on this? are you working with anyone? fwiw i still find it a bit chaotic and have some ideas about the UX, like i think having a unified UI vs. a bunch of different terminals, e.g. would be really helpful. i have to take some time and see what they actually wrote, i couldn't even really keep up with how fast they were going, but it's just really impressive, and i'm amazed at how close it comes to feeling like they are communicating in real time. the context window is another area that i think could possibly be improved, i had to kind of bop around and opportunistically compact when i found a good chance - but i'm sort of curious to see what your actual workflow is, do you stop and start new agents a lot, do you just let it auto-compact, etc.?. anyway please let me know if you have any thoughts, i definitely want to share this on the list, and i'm happy to share it as it is, which for me is a bit rough and wild, but you done such a good job with it that it's worth sharing either way.
also it looks like you're still doing active development which is great, and for what it's worth if you are interested in collaborating a little, i'd love to be involved in this project, but you should let me know what works for you, otherwise i'll just take some time and read through the code a bit more carefully and stuff, and let you know if i have any concrete suggestions, if you're open to that. 👍🏽 reallly cool stuff
EDIT: btw if you want to chat, you can also hit me up at my new inbox: hesreallyhim/my-pgp-inbox - it looks like you have a gpg key so it's supposed to be kind of a public-private DM thing using PGP, it just set it up so i'm keen to play around with it. cheers.
EDIT: Sorry i also was curious what did you use for making that screenrecording in the README? i've been working on some stuff with vhs lately and i'm just curious.
OK, so I don't know if you saw my previous comment from a while ago on the awesome-list - the first time i tried this, i liked it but it felt like really hard to control and things were running off the rails and i was a bit overwhelmed/confused. to be honest, i still kinda feel that way a bit lol - maybe it just takes some practice, but i tried it out again with the latest version on a small greenfield thing i'm playing around with, and it's really exciting to be honest - prob the most interesting thing i've tried in a long time. i had 3 agents going and they're communicating with each other, and doing a pretty good job of not getting lost or stepping on each other's toes, correcting course if they get mixed up (ok i'm using Opus so maybe this helps), and just got a ton of stuff done so quickly. i was hardly able to pay attention and it was a toy app so i kind of just let them go, but they seem to do an amazing job of coordinating things. (I think, at least, i haven't gone back to read the code yet.)
how's progress going on this? are you working with anyone? fwiw i still find it a bit chaotic and have some ideas about the UX, like i think having a unified UI vs. a bunch of different terminals, e.g. would be really helpful. i have to take some time and see what they actually wrote, i couldn't even really keep up with how fast they were going, but it's just really impressive, and i'm amazed at how close it comes to feeling like they are communicating in real time. the context window is another area that i think could possibly be improved, i had to kind of bop around and opportunistically compact when i found a good chance - but i'm sort of curious to see what your actual workflow is, do you stop and start new agents a lot, do you just let it auto-compact, etc.?. anyway please let me know if you have any thoughts, i definitely want to share this on the list, and i'm happy to share it as it is, which for me is a bit rough and wild, but you done such a good job with it that it's worth sharing either way.
also it looks like you're still doing active development which is great, and for what it's worth if you are interested in collaborating a little, i'd love to be involved in this project, but you should let me know what works for you, otherwise i'll just take some time and read through the code a bit more carefully and stuff, and let you know if i have any concrete suggestions, if you're open to that. 👍🏽 reallly cool stuff
EDIT: btw if you want to chat, you can also hit me up at my new inbox: hesreallyhim/my-pgp-inbox - it looks like you have a gpg key so it's supposed to be kind of a public-private DM thing using PGP, it just set it up so i'm keen to play around with it. cheers.
EDIT: Sorry i also was curious what did you use for making that screenrecording in the README? i've been working on some stuff with vhs lately and i'm just curious.