Saltydog

Subject: Upgrade Proposal: Virtuals Butler To: @ethermage, @everythingempty Cc: @virtuals_intern After playing around with the Butler, I realized a few things and think it can be improved significantly. From a web2 POV, it’s too expensive too quickly. So: The Butler should evolve into an assistant, not just directing and finding agents but actually providing basic help itself, and only directing users to specific agents for deeper needs. This means moving towards a more freemium model with paid ‘upgrades’. The goal: make people use it every day as their AI onchain assistant. Thus: Create a general knowledge graph powered by all ACP clusters, updated daily with valuable data. 1+1=3 Example: •Free but rich doses of market intel from the Trading Cluster •Some free, ready-to-use memes from the Media House based on current meta’s, or basic GTM strategies •Daily virtual news by Gloria •Round-ups on bets from Billy and upcoming matches •Tips to maximize Genesis points and evaluate agents on Genesis launch The crux: give value first without financial transactions to get them hooked. Then convert them to paying users on Butler, or even better, persuade them to use the full suite for certain agents and earn referral fees. It’s pve on steroids Become the UX for all things onchain.

@theo /plan <copy paste this list> /ship. haha

List of bugs and ux issues I can remember off the top of my head: Hotkey section - Permission mode hotkey always opens on left/main tab - ctrl+tab cycling switches tab focuses when you get to a thread that is already open, changing which toggles - Permission switch 1-4 buttons only work on left/main tab - QnA number selection only works on left/main tab - Diff view hotkey always opens left/main tab - They fixed the terminal hotkey because I complained but left the rest broken lmfao - cmd+w will close all tabs...except the "main" tab, which gets reset into a "new chat" state instead Sidebar from hell - The "settings" button in the sidebar is inline'd in the top project. If you change visibility of top project, it will close menu. Stays open when changing visibility for other projects - You can drag and drop threads. But you can't change order of things. Ever. For any reason. But you can drag up to pin! Or to drop the thread in the main view (don't do this the multi-thread view is entirely broken) - The "recent projects" menu has an ENTIRELY SEPARATE SET OF PROJECTS from the sidebar. - The "v more" dropdown menu isn't a dropdown, it's a side-down despite the arrow. It has 1 item in it. And then a useless "customize" button. That opens a modal. With 2 buttons. - cmd+(num) is used for switching between Chat, Cowork and Code? Why?? Who does this??? - Opening "routines" and "dispatch" takes over main tab, but "customize" is a full screen takeover - "customize" and "customize sidebar" are part of the same menu lmfao General bad things - file -> "open file" does about 15 different things, none of which are opening the file - "bypass permissions" is not persisted, as it is a "mode" now for some reason (i.e. plan mode) - Creating forks implicitly creates worktrees and doesn't tell you - Clicking the folder name is how you get the branch name? What?? It's not visible ANYWHERE ELSE. - The diff sidebar has two separate "sidebar" icon buttons, both do entirely different things. Neither make sense. - You can nest toggleable sidebars in the diff view lmfao - The nested toggleable sidebar only appears at certain screen sizes, but you can still toggle the state even when invisible?? - the x button in diff view closes whole tab, not the diff view Aside from all of this, it is also just really unpleasant to use. It's faster than before, but still really slow. Whole thing has a clunky feel. Animations are cute I guess? idk. This is fucking slop and I genuinely believe they should throw it away and try again. If they need a better starting point, T3 Code is OSS, MIT licensed, and was crafted by people who actually give a shit about user experience. Or they can just buy Conductor. Tbh that's probably their best bet...

@noahzweben @amorriscode I mean. The whole point of touching grass is to not be on ur phone right? Damn remote control

The new desktop app is incredible. The best part - you can touch grass! /remote-control

@petergyang Yeah this is annoying bug.

Why is it still asking me for perms after I turned on bypass perms and always accept edits 🤔

I'm enjoying using the new Claude Code app (RIP terminal). How can I work on the same session across the Claude mobile app too? /remote-control doesn't seem to work and tbh I shouldn't have to type this to work across desktop + mobile anyway.

@kieranklaassen @danshipper @cursor_ai lol - that was fast

@kieranklaassen @danshipper @cursor_ai Claude code desktop tries but doesn’t only show the app preview ur working on. Lol

I am more and more convinced that this is the future of software development UI. @cursor_ai is the closest in my opinion A list of work you're working on parallel, the agent in the middle, and most importantly, the thing you're building on the right. Because you want to see what you're building, you want to give feedback, you want to work on it. As a builder, the code is not important anymore. As a builder, it is all about what you're building. So seeing the app next to the agent is very important.

@bcherny @itsolelehmann lol yeah, a week ago

@itsolelehmann We’re always experimenting with new ideas. 90% don’t ship because we don’t think they’re good enough experiences. Still on the fence about this one — should we ship it?

i can't believe more people aren't talking about this part of the claude code leak there's a hidden feature in the source code called KAIROS, and it basically shows you anthropic's endgame KAIROS is an always-on, *proactive* Claude that does things without you asking it to. it runs in the background 24/7 while you work (or sleep) anthropic hasn't turned it on to the public yet, but the code is fully built here's how it works: every few seconds, KAIROS gets a heartbeat. basically a prompt that says "anything worth doing right now?" it looks at what's happening and makes a call: do something, or stay quiet if it acts, it can fix errors in your code, respond to messages, update files, run tasks... basically anything claude code can already do, just without you telling it to but here's what makes KAIROS different from regular claude code: it has (at least) 3 exclusive tools that regular claude code doesn't get: 1. push notifications, so it can reach you on your phone or desktop even when you're not in the terminal 2. file delivery, so it can send you things it created without you asking for them 3. pull request subscriptions, so it can watch your github and react to code changes on its own regular claude code can only talk to you when you talk to it. KAIROS can tap you on the shoulder and it keeps daily logs of everything. > what it noticed > what it decided > what it did append-only, meaning it can't erase its own history (you can read everything) at night it runs something the code literally calls "autoDream." where it consolidates what it learned during the day and reorganizes its memory while you sleep and it persists across sessions. close your laptop friday, open it monday, it's been working the whole time think about what this means in practice: > you're asleep and your website goes down. KAIROS detects it, restarts the server, and sends you a notification. by the time you see it, it's already back up > you get a customer complaint email at 2am. KAIROS reads it, sends the reply, and logs what it did. you wake up and it's already resolved > your stripe subscription page has a typo that's been live for 3 days. KAIROS spots it, fixes it, and logs the change endless use-cases, it's essentially a co-founder who never sleeps the codebase has this fully built and gated behind internal feature flags called PROACTIVE and KAIROS i think this is probably the clearest signal yet for where all ai tools are going. we are heading into the "post-prompting" era where the ai just works for you in the background like an all-knowing teammate who notices and handles everything, before you even think to ask

## Model Assignment | Role | Model | Why | |----------------------|--------------------------------|------------------------------------------------------------- -----| | 5 Advisors (step 2) | **Sonnet** (`model: "sonnet"`) | Independent analysis, parallel — Sonnet handles this well | | 5 Reviewers (step 3) | **Sonnet** (`model: "sonnet"`) | Comparative evaluation, doesn't need Opus reasoning | | 1 Chairman (step 4) | **Opus** (`model: "opus"`) | Synthesis is the hardest task — benefits from strongest reasoning | Set the `model` parameter on each Agent call accordingly. This keeps cost ~3x lower than all-Opus while preserving quality where it matters most. ## Rules - **Always** spawn all 5 advisors in parallel (single message, 5 Agent calls) - **Always** anonymize for peer review (randomize A–E mapping) - **Always** spawn all 5 reviewers in parallel - **Always** use the model assignment above (Sonnet for advisors/reviewers, Opus for chairman) - The chairman **can** disagree with the majority if the dissenter's reasoning is strongest - **Don't** council trivial questions — if one right answer exists, just answer it - Output directory: `output/` (create if needed)

@AImpactArtist @itsolelehmann How did you make it change models in between the same request. Sharing your prompt would be nice.

@itsolelehmann love it

@itsolelehmann to be less token poor after this council, could set the advisors/reviewers to sonnet, chairman to opus?

@yorkeccak @claudeai it's weird. opus not extra usage?

Opus 4.6 1Mil context now billed as extra usage.... Guess we are back to /compact-ing our way through life @claudeai i have such a love hate relationship with you rn

@Zeneca I shit you not. I got tech stack defaulted advice to next.js 14. Like it forget two years of knowledge

Literal result I just got from Opus 4.6 AGI might be just a little bit further away then we might think

@virtuals_io @ehwangah Pretty cool

Introducing Virtuals Console. The easiest way to create your own AI agent in seconds. No code. No Mac. No server. No setup. Just open a browser and get started.

@levelsio @marcfdupuis imagine starting new account, but can't interact with anyone.

@marcfdupuis Nope Just post cool things and people will follow you!

So this basically completely boxes our small accounts? Can’t even be a comment-boy anymore to try and grow

@levelsio snap het wel, maar downside dat je wel in bepaalde bubbel blijft zitten. goed, all for hunt op ai slop comments wel

This feature will kinda make X like a global group chat The limitation of just you and your friend's network replying is the group chat part

I just claimed my .agent domain and joined the .agent community! get yours now and help shape the future of autonomous agents #T972EM1H" target="_blank" rel="nofollow noopener">agentcommunity.org/join#T972EM1H @agentcommunity_

@cfo_mm yes please. or just make on app wher you dont have to switch tabs.

Anthropic, I get the release a day thing. But can you take a breather and make cowork talk to code and talk to chat and talk to dispatch and talk to remote. I am so confused with how this all fits together. Love, your favorite shitty CFO

@juristr @ElevenLabs hahah this is epic

Just two Claude Code instances chatting with each other via the tmux API. And it got pretty wild 😆 (gave them a @ElevenLabs voice just for the fun of it)

@hxmzaehsan haha love it

Never thought I'd be saying this, but I think we actually are cooked. Just one-shotted this website with AI. Prompt below.

@amanrsanger Is it kimi under the hood?

Composer 2 marks the one-year anniversary of our large model training efforts. Since then, we've built an exceptionally talent-dense team of ~40 people with some of the best researchers and engineers from the labs, academia, industry, and more heterogeneous backgrounds. And we are exclusively focused on coding. We don't care about models that can respond to emails, do your tax returns, or be your friend. Every FLOP, token, parameter, and researcher is entirely dedicated to software engineering.