Introducing Agentbox, the new meta for working with agents. Human attention is the new bottleneck of AI work, and Agentbox lets one person manage 20+ at once. - Every agent chat window collapses into a single inbox - Agents stay quiet until they need you and prioritize themselves -... Lightning fast, keyboard driven, and built for flow Free for Mac, open source, and runs on your Claude Code or Codex plan.show more

Download free:

human attention really is the scarce resource now

Indeed!

Making interrupts asynchronous is the right call. The tricky part with twenty parallel runs is usually git collisions and state overlap—does each run in its own worktree?

Yep, each has its own worktree!

this is super cool!! @GaddipatiHarsha bro get @slashyemail on this

@GaddipatiHarsha @slashyemail Thank you mon amie!! And beautiful product you tagged!!

Just don't trademark the name because this is what my vm is called lol

Don't plan to! Great minds ;)

Wait this is cool

Appreciate it :)

Stoked to give this a try

Would love to hear what you think!!

it the bestest way possible, feels SO superhuman. I really think this has so much potential. Will screen record.

This made my whole entire day. You’re a legend Bud

I built helix, a personal tool to manage my AI agents, so this is immediately my kind of problem 😂 the quiet-until-needed part sounds especially nice.

Congrats on the launch! This looks sweet

Thank you Mr Rich!!

Nice 👍 job on agent box 📦

Thanks :)

“Every agent chat window collapses into a single inbox” is what I’ve been wanting. Some centralize place where my agents can send notifications to for me to approve/deny

You get it!!

This is sick! Ever consider running this on Apple Messages for Business so people can text via their iphone?

Interested in the idea, how are you thinking it'd work? Def planning a mobile app

Your users will be able to make decisions from Messages on their iphone - AMB has the ability for richer components like list pickers, menus, etc, so if you really want to avoid the chatbox interface, it's possible. Or you can text it, up to you. Check out grab a meeting, and we can walk you through it! No commitment obv. Or you can dm me

stoked to try, looks great!

woah I actually really like this

Yay! Appreciate ya :)

Launch queen is back with another banger!!!

Content king is back with more support!! ❤️

Congrats!

Appreciate it!!

@doodlestein agent inbox.. sounds familiar? 😉

@doodlestein Tell me more!

spiritually similar to Jeff's "MCP Agent mail". which is a key peice of his "Agent Flywheel" system that allows him to ship like 5k PRs a month or something crazy. The diff is that it's an inbox to let agents talk to each other, whereas yours is about human in the loop (correct me if i'm wrong). Either way, it's a neat project. I just don't know it's for me personally!

biggest fan rh!

That's crazy cause I'm YOUR biggest fan!! Can't wait for our ice cream catchup

LY!!

🫶🫶🫶

This could revolutionize agent management

+1 it certainly has for me :)

one finished task beats twenty agents asking for your attention

quiet until needed is the key

Makes you many times more leveraged

OSS FTW!!!!!!

TAHA FTW

🤯 Smart

20 agents only works if the 'needs you' classifier has a low false-positive rate. what's the actual interrupt rate per hour once your inbox is full, and who gets paged first when three agents block on approval at once?

False positives only cost a few seconds, and when agents interrupt incorrectly you can just adjust the system prompt. They organize in your inbox by urgency. In practice I run 20+ threads on default settings without issue

Self-prioritizing agents, hope they don't all need you at once

That's the beauty of the inbox - it's designed for batch processing. Every morning I've got a whole load to process from the ones that worked overnight, and it's very enjoyable

inbox collapse is smart but agents staying quiet until they need you is the hard part to get right. most will interrupt too early to feel useful.

You’d think so, but it’s actually been a non-issue so far! It takes seconds to process an agent that returned early, and you can update the system prompt so it doesn’t happen again. + you can design the system to replace HITL steps with verifers At that point the bottleneck becomes memory on your computer to run agents in parallel, but we're working on that!

so cool!

Thanks Benji :')

One inbox for 20 agents sounds great until they all need you at once lol. How does it decide who goes first?

Love this take

we like this

I am not really sure this is worth a separate app to download on Mac. I have been doing that within agents and artifacts for ages though. Still I am glad you are putting it out there for people! 🙏

Thanks! Totally understandable, desktop app is largely so you can use your existing Claude Code and Codex subs. If friction is too high you can also run it in the browser with one command: npx agentbox-app

@Grok 给我介绍下这个