v2 has some nice stuff with background subagents and shell commands

note beta is out

this is good, i was wrong

i didn't understand bg work for a while because the demos kept showing "run vite in the background" which i didn't care about but then i understood it's more about unblocking the session when it was annoying

Having squatting rack behind my desk is my new life goal

Is there a way to register subagents to a specific model or is subagents a task tool tied to the main agent?

yes you can, ask opencode how to set it up

i love this app ever since i first used 5 months back

nice stuff on WrongStack 😎

Very cool

can i buy that hat

so is gpt able to spawn in a grok agent if i js said spawn a subagent w x model

yeah ask opencode to configure a subagent that uses grok

why not make this like cursor's though, it has to go into opencode.json & add in custom subagent entries its not truly an "ok opencode spawn in 3 subagents make one use meta spark 1.2 & ox alpha & mimo v2.5" cursor has this

@thdxr yeah we are working on how to do this

@thdxr THANKYOU MR REBASE

curious how this is implemented. when you background the process, are you cloning the existing session and pushing the prev. session to background? the obv. step would be to probably spin this into a background subagent (but i'm curious how the context is being handled)

For the love of god, please make the model + reasoning variant a parameter when invoking a subagent in v2🙏🙏

Very cool

I think i’d be great if you could share some productive sessions where you used this. I’ve tried it by running the v2 branch 2 weeks ago and felt quiet bad to me, but I feel like this may just be me.

idk if we had issues back then but i don't know if i have much more to say than what i said in the video 1. i use it to unblock the session when i want to move on 2. sometimes during the session i think of something tangential and i'll say "btw in the background look into why those tests are failing" while i continue with work can you explain a bit more what felt bad?

I just saw your comment that it's about unblocking which makes total sense. I am currently trying to go from pi to opencode and have been struggling with subagents and what they did in general. Probably also cause I tried the orchestrator.ts. but than it's a diff topic thanks!!

When do we get a real review screen in tui?

We need subagents to be able to delegate to smaller models on their own though. Shouldn't do it by default but it would be very powerful to be able to ask to do this. (A big model like 5.6 sol could do exploration with smaller models like 5.6 Luna, DeepSeek V4 flash, etc.)

@thdxr you are such an inspiration dude. One of your post resonated so much with me where you talked about building things even if it was already built by someone else.

we need i18n

For tasks moved to the background, is there a notification upon completion? If hallucinations occur, how can I tell?

Nice, yeah I'm always berating my clanker for hopping into like a 10 minute build process or something when I was still chatting about tactics/next steps. so this is super handy 👍

ok but like when is it out

yay

very interesting! so if a current task (i.e. build command for your app) is running and its taking a while. Do you move it to the background to work on something ELSE or is it continuing running other prompts? I would like to see a real world case of how you would use it? Most of my tasks are quite single threaded but I might be able to learn how to think is a multi-threaded way if you could share few examples!

Bruh and I build an async task plugin for opencode. (Will still use it as other mcps can throw back more and more overtime & it allows for watchdog behavior)

@opencode Sweet

awesome sauce

curious how you're handling the case where a background subagent runs a shell command that changes state the main session assumed was still clean, is there any kind of lock or does it just hope for the best right now?

Cool but does it have vim mode?

Interestingggg. This got me thinking. I wonder what the experience would be like if all shell commands automatically behaved like this

background subagents are the interesting bit. shell commands make them practical.

sick, keep me in the flow baby

sleek! btw what happened to the todos tool and the sidebar panel integration of it? :( really missing that one!!

can the agent decide for itself whether to spawn something in the background vs blocking? that would be the ideal for me

🐐

Sick!!!! YES! When we getting it?

Wow looking intelligent

Are you guys going to implement dynamic workflows? I implemented one myself but I think if it is natively supported you guys can make the UI UX way better