Hacker News Reader

One thing I don't understand with all of these, is how they're handling different worktree infrastructure spin-up?

For example, if I have a webapp, I want each of the worktrees to spin up its own infrastructure, and be accessible on its own unique local url, so that I can see the changes locally for each worktree, or I can have agents automate visual checks using something like agent-browser.

Currently I use docker for my infrastructure, each service running in its own container. I have a script that has a ./app worktree create worktreename. That creates a worktree as "worktreename" and spins up all of my docker infrastructure with prefixes for things like "WORKTREENAME", and I can access all my urls at worktreename.myapp.test (or just myapp.test for the main worktree).

This is working fine for now, but it'd be cool if one of these apps was compatible with this concept so I could move over to that.

You just enter a setup script that it runs on worktree creation

"just" is doing a lot of heavy lifting here. I always automate my repo setup, usually with mise tasks, but there's often assumptions baked in, eg around which ports the app will listen on. It's also still hard to get the agents to use "mise run test" instead of invoking pytest directly. This sounds harder to deal with when you have multiple copies of the test or dev stack running together on the same machine.

I mean I'm not sure what your issue is. There has to be a setup script of some kind. What are you saying, that your setup script is too... dynamic?

this is not hard

I was looking exactly for something like that. I tried installing the Linux version and it runs but:

1. it's behind a login wall | 2. tries to download its own OpenCode instead of using the one installed on my machine

I also tried to create a new workspace. It asked what I want to do. I tried with "create user accounts", and it proceeded to create the git worktree (without asking for its name) and sent that prompt straight to OpenCode without allowing me to choose the LLM model.

I plaud the effort - I really do - but it doesn't really seems a great experience for now.

Wow - this looks like a fork of my project https://github.com/frenchie4111/harness

It's crazy to see how we have independently landed in the same place

Good luck to your project! Excited to see how our projects differ in the future

Amazing, this is exactly what I've been looking for a while now, especially the remote dev box part. I'll give it a try, thank you!

Remote dev box is brand new in the last release so let me know if it works... happy to help you set it up (I personally run w/remote dev box) shoot me an email

looks awesome! very interesting how similar these tools all look. i do think we're at an intermediate state here with the current UI where it doesn't scale as well for the future where triggering agents and review becomes much more important than the writing code part.

Agreed - On Harness I am already working on triggering/review. Those are the two next projects on the list

We will definitely continue to iterate / try to keep up with whatever everyone thinks is the best UX

Looks great, even has Linux support!

And headless server support! I have gotten a lot of request to let people run the agent on a separate dev box (in the cloud or some pc gathering dust in the corner). The frontend/backend are decoupled so you can run the agents on a totally separate machine than the GUI is running.

If you have any other problems/concerns/issues/suggestions/etc, reach out!

That still is not it for me, for now iTerm2 tabs / tmux is plenty. This, or tools like Warp, feel so heavy for me and I get overwhelmed.

The Codex/Antigravity desktop app alternate route feels more like the direction I can buy into it, but I still feel like there's room for another novel UX not yet done.

Perhaps the biggest thing I feel like I wish existed is a "Scratchpad" mode inside a session, Claude recently came up with `/btw` and it's been really useful. Same with `!` for terminal commands within session.

Nested thinking spaces within a chat window UX is a very tricky thing to reason about.

yeah us and similar categories of tools prescribe a lot of our specific workflow for worktrees, review, code editing where iterm + tmux is much more flexible. we have to tow the line of what's useful as a built-in feature vs being very agnostic.

i do agree that there's a more novel UX than the same left sidebar with worktrees and middle chat/terminal that is not yet explored.

we've been spiking on what the UI is like to program and monitor the specific steps in the workflow from triggering agent -> planning -> coding -> review -> merge.

I've actually was reading up pi.dev website copy today and something stuck with me with what I think is the direction that everyone is lacking.

They say "Change the harness, not your workflow". I think that captures what nobody is doing it right, I feel like there's a large effort towards over-engineering for a workflow and assuming it to be optimal when the real lever to press should be the harness.

Anyways, just an idea, wanted to throw it back your way since you're building in this space. Good luck!

I've been using this for the past few months, and I love it! It's built exactly around my workflow with many worktrees in various repos open at the same time, sometimes with different agents working side-by-side. Before Superset I just used terminal tabs but simply couldn't manage more than like 20 terminal tabs without losing track, so i coudn't scale further. Now i'm running probably 40-50 agent sessions over several repos simultaneously without any issues and losing track! Keep up the good work guys!

Could you elaborate a little on what you are doing with 40-50 agents? I use Claude, I've employed sub-agents, but I still can't wrap my head around how people are using them to that extent.

ha, I didn't mean they are all running at the same time. sorry if it sounded like this. I open a new worktree/agent session for each new feature or bugfix. Usually I start with ideation and brainstorming before doing the actual implementation. However my priorities shift daily, so I could be starting a feature on monday, but then stop during the ideation already, pick up another task on tuesday and won't go back to the initial one until a week or a month later. Since I got it in a neat worktree in superset, I still see my open branches easily, have access to its claude code session and can resume the work much easier than before having terminal tabs. Because of this setup I currently have 17 backend sessions, 25 frontend sessions, plus several other repos open and ready, when i need to.

Ahh got it, thanks. That makes a lot more sense -- indeed I thought you had 40 active, parallel sessions!

awesome to hear this! 40-50 is definitely on the high end. Are you adding any workflow on top to manage that many?

just gstack which comes with shared memory, so each session has access to another's state if needed.

ah very interesting will have to give gstack another try

glad to hear it! thank you!

one issue i'm having, maybe you know what's causing this, in long running terminal sessions I'm getting rendering issues. Like some characters show as something else. I have to resize the window to fix it, usually opening and closing the code sidebar fixes it for some time, but it never fully goes away. I think i also have it more on the big screen (4k) than on the laptop screen.

ooo this could be a webgl context bug, we are working on fixing this @keit

Nice. In the right track. I made something similar, but focused on local agents, but we both have issue tracking for managing multiple project and agents in parallel. It works, I think people will be surprised when they start using systems like this.

It is very different from current editors and the direction they are going in. In a way, it undermines the direction they are going. Current editors aim to make engineers 10x or 100x. These editors aim at a different target than the engineers. I will leave it to the imagination on who.

thanks! yeah we daily drive superset so it definitely has been working for us, and yeah these tools are gonna end up looking pretty interesting :)

Nice if didnt have the same name with an Apache project https://superset.apache.org/

Why Mac only?

Also - one issue I've seen with other tools doing worktree stuff is they don't deal with merge conflicts automatically. IMO the agents should just automatically resolve conflicts & rebase on their own, is that a thing here?

We technically have a linux build, but dont maintain it as well as mac. We are built on electron so eventually we want to more heavily support linux and windows but we dont currently have anyone on the team that daily drives linux and windows the maintenance overhead is a bit too high. Once we do have someone on the team driving linux/windows we'll have broader publicized support for all platforms.

And on the worktree note, I find when working in a worktree the agent has a much easier time solving merge conflicts and as the number of feature branches scale and it just makes it so i dont have to worry about conflicts while working on a feature.

Surely that's something you just instruct the model to do in your CLAUDE.md / AGENTS.md right? Not really the domain of the IDE.

If you try to just instruct them they might get it wrong. If the surrounding software forces them to do it, it'll always work. E.g. it can check for merge conflict markers like <<<<<<< and re-invoke codex/claude to merge again if the previous resolution failed for whatever reason (e.g. AI hallucinated, threw up, whatever).

Also you'll need a wrapper to actually detect when merge conflicts will occur and when rebasing is necessary.

Generally, the less you rely on the AI the better. Make the AI write the code, sure, but don't make the AI be the process.

I think that's being pretty uncharitable to the model... I use Opus and it has zero issue handling this

I always feel like I want to use something like this, but then realize my NeoVim setup + tmux + Ghostty is good enough and I am not ready to learn a whole another system for modest gains.

The friction I have currently is obviously things like port forwarding, in app browser etc.

I keep thinking to try out cmux but haven't gotten around to it.

makes sense, well if you ever want to dabble, let us know how you like it :)

I used Superset for quite a while until a month ago. There were some annoying issues, with freezing and terminal not being rendered how it should be. And they did repeated fixes that didn't really solve it. Since I had work to do I moved on.

I installed Zellij on my server where most of work is happening and local machine and this works well for me. There are other issues I have now, but overall flow is fairly natural to what I am doing.

I liked that they did integrate a lot of agent workflow in Superset but my experience was that it would just take too many resources and especially with glitches, it wasn't worth it continuing. I had a period where i enjoyed working in it. It is vibe coded electron app, 2GB! is too much for this kind of app.

I just updated to their new version... it supposedly imported my projects but I can't find anything... so... I guess this is it.

sorry to hear about the issue. we really messed up on the performance and balancing that with more features. looking into the imported projects did no projects show up on the sidebar for you?

will continue working bugs and hope you'll give it a try later in the future when the product's more stable :)

Just switched from Conductor to Superset and I'm a big fan. I really didn't like the extra stuff conductor added to the UX (the text rendering always drove me nuts).

So far so good with Superset - even as a non-engineer builder.

Glad to hear it! :) we'll keep shipping and making improvements

Personally, IDE for the agent era is just Linux.

Kitty with oh-my-zsh, lazyvim and an agent. The entire thing is an ide. If I need to refactor, query data and interact with the system I just use native tools like rg+fastmod, bash, awk, jq... Either writing myself of asking an agent to do the heavy lifting.

Linux in the agent era is a breeze to operate and reason about, so the whole thing becomes a single development environment that's really light on resources and effective.

What I've started to do is to use Zulip. You can have different agents in different chats. You can upload files and you update from your mobile phone. At first I thought it was crazy but it's nice not to have 3 different AI agents running in tmux

I am looking for something similar. Any tutorial or resource on how to set up things?

sorry, I'm back. Please feel free to email me at {myusername}@gmail.com. I'm glad to share with you the python code that Gemini created for me and then I asked Claude to duplicate it for Claude.

A simple GitHub gist would do the trick, just pass it through gitleaks to see if you aren't exposing secrets and you are good to go.

That's an interesting take! Basically Linux / a computer is everything you need to ship code.

If I could provide one gentle pushback - the same way there's utility in OMZ, lazyvim etc., there may be utility in us shipping our CLI etc. - there must exist some software we can build that'll be useful to you as well :)

I've been using Superset as my daily driver for about 2 months and I'm loving it.

i have created a simple setup script that reserves a couple of ports per workspace/worktree. with that, each ticket/workspace can starts it's own webserver. Allowing me to work on multiple things in parallel. that also allows me to use Claude+Chrome MCP to e2e test each implementation in parallel.

Besides my own repo I also have a couple open source repos open. so I can quickly ask Claude questions about it instead of looking at the docs, which are usually outdated.

All in all, I usually have ~5 workspace/tickets/agents running at the same time. And one as Q&A agents per OSS repo

At first glance, it looks similar to Conductor (https://www.conductor.build/). It seems like a lot of these tools are converging on the same general ideas.

Could you share a comparison with the other tools out there?

For me the greatest difference is that superset is terminal-centric while conductor is chat-centric.

GUIs slow you down, in my opinion. But having the nice visual diff is something we can't really do well in TUIs, so very welcome.

So, superset, for me (been using for quite some time now) is basically to organize my agent and terminal sessions per task and project.

I can switch context much easier and can also resume working on something days later, with all my tabs nicely available and separated.

This was consuming me before, a dozen or more tabs and windows in my computer that I don't really remember to which task each belongs to.

Note that Anthropic specifically called out that usage through Conductor will be metered as "programmatic usage" in their June 15th pricing change: https://news.ycombinator.com/item?id=48126438

My colleagues are planning to get around this by using Conductor's big terminal mode. At which point I think it's basically just a git worktree manager?

If you want to do that you mine as well give Superset a try, we've put a lot of time into giving a good terminal experience, with baked in notifications and a daemon that runs in the background so your terminal sessions dont die when the app updates or closes.

Nice work Avi, you got a user out of this message, at least for now. The background service keeping the terminals alive is lovely stuff.

I don’t see any specific mention of Conductor, is this confirmed?

yeah there is a lot of overlap, we are more terminal first than conductor so you can do can use any cli agent you want. We have a lot more quality of life features around the terminal like notifications, and some things similar to tmux where if you kill the app or update your sessions stay alive and running. We also recently released remote workspaces so you can setup cloud workspaces for your agents. Id say if you like the chat experience conductor is still a bit more polished, we'll get to that level of polish soon, but if you care more about the terminal and cloud and more new integrations we are shipping superset is better.

Very excited about the remote workspaces! Perfect for me as I've set up a remote Mac as a Claude terminal. I know it's only just released but the typing is quite laggy... have you considered something like mosh? I've been using this from the terminal and it makes it feel as fast as running locally.

When you say “terminal first”, are you terminal-first enough that I could use vim buffers for editing?

yes, it's essentially a terminal with extra agent hooks tracked and more customized review flow. if it runs in a terminal it runs in Superset.

What's the difference between you and emdash and conductor and t3

Conductor and t3 - We don't build on top of the SDKs and don't plan to. As a terminal first we give you the flexibility to work between the different CLI agents CC, Opencode, Codex, etc. to get the latest features.

Emdash more similar but we're planning on investing more into allowing you to script and automate the app + agents more with our CLI.

Sadly there are so few solutions, if even one, that is trying to offer a real remote agent interaction. That's why I've build my own. No IDE per se rather than a discord bridge, with all interaction possibilities that makes sense and a way for my agent to quickly send me files or host mockups. My usecase is to build tools, reports, do research, mock up ideas and build prototypes. That's why I don't need to see the code that was written.

hmm well we do have a slack integration and forsee a discord integration in the future so maybe we'll end up getting there for you :)

Loving superset! Congrats on the launch! Excited to try remote workspaces so I don’t have to leave my computer open while I have autoresearch running.

Thanks! glad to hear it :)

What are you doing with multiple agents, i.e. how do you have so much work to do? I can't come up with new ideas faster than the dude is coding.

there's a lot of cases that works for me with multiple, spiking or triaging issues generally take long so i usually can kick off a few different workspaces for to check back in after a while. what i don't want to do is wait around for minutes while an agent is coding or exploring.

Is anyone actually using agent swarms for anything real?

Yeah, how many agents can you people even run at once and how much does it cost you? In company we used the monthly token quota and nowadays it's basically unusable with claude opus 4.6 on high reasoning. You can basically burn through 100% usage through a single day. How does it even scale for you with N agents and which magical plans or models do you use, where tools like this are even viable?

for clarification we're less of a agent swarm tool, and more of a launch a bunch of independent agents in isolated environments and have some nice UX to manage them too. I also havent had as much luck with agent swarms or ralph loops, but i'm sure the laps will improve them with time

How do you guys plan to sustain the business, given that your product here is open source & already has many competitors doing similar things?

So far we've been growing pretty healthily all things considered!

I think one thing to remember is that the other side of us having dozens of competitors is that if the space couldn't sustain more than 1-2 parallel agent companies, a lot fewer of us would exist. We also will have a lot of time to continue creating value for our customers in the future in new ways :)

we monetize on teams and, in the future, cloud. the bet is that teams will want to centralize their set up for this type of work, especially shared Linear, GitHub, skills, etc.

The FAQ says "Superset has a free tier. The source code is available on GitHub under Elastic License 2.0 (ELv2), so you can inspect and self-host it subject to the license terms." - what is self hosting in this context, isn't it a desktop app? Is this why it wants me to sign into something? What exactly am I signing in to?

So we also ship a cloud service along with Superset, which enables our Linear integration, Slack integration, and our multiplayer capabilities / remote workspaces.

When you sign in, you're signing into our cloud service!

Got you, thanks. So if I don't need those things I can skip the sign-in?

Ah unfortunately not through the hosted product, but if you fork and build it you should be able to skip the sign-in

I'd love a comparison to what's already out there. Don't vscode, antigravity, cursor etc all have agents too?

Yes. Antigravity switched to primarily be an agent management tool (the previous version of the product became Antigravity IDE). Additionally, most advanced tools automatically spawn subagents.

biggest difference is it's terminal first, and optimized for CLI agents. we don't prescribe a specific harness and instead try to work with any CLI harness you bring.

I agree with the hard part being managing state, especially environments and ports. I've never used lsof so much in my life.

Question on Remote Workspace: Can the remote machine port forward so I can use a browser to see / test current state of the app on the remote machine?

On the docket! Right now the main thing we have enabled is the file system + terminals + ai agents through remote workspaces, but yes dev environments is definitely on the agenda :)

I switched over to Superset from Conductor a few months ago and haven't looked back - it's really nice to be able to use the native Codex/Claude Code TUIs without any of the bloat

Can't wait to see what else you guys cook up!

That’s awesome to hear!

Definitely some exciting stuff coming with better automations, mobile and remote workspaces

Glad to hear it thanks for the support!

Sorry if I'm missing something but is there a way to create a new Claude workspace that has dangerously-skip-permissions but starts in Plan mode? There's no mode selector in the new workspace modal

Yeah I have --dangerously-skip-permissions configured as a preset for Claude in Superset, but you can add more commands too. Just go to Settings > Terminal

That overrides plan mode though, eg it seems claude --dangerously-skip-permissions and --permission-mode plan can't be used together. So I can't kick off a workspace from Superset with a prompt that starts in Plan mode but can switch to dangerous

Ohh I see what you're saying. I guess you can start in plan mode and then once you're ready to tell it to proceed then shift+tab to auto mode?

is it terminal on steroids some kind of? so you can manage mutiple coding agents? how many coding agents you can manage in parallel that it is still comfortable to work and code changes are meaningful

yes, we surface agent states automatically so you can see what's running or needs attention across the different workspaces. there's a set of tasks where having 5-6 running in parallel is still productive for me such as running spikes and fixing small issue.

As we're investing more into integration test and self-validating for the agents we're able to increase the number without sacrificing quality.

This uses separate git worktrees. If we have a local dev setup involving multiple docker services, is there a recommended solution for managing those envs? I didn't see.

We have a concept of setup and teardown scripts if you're interested in checking them out! Together with worktrees, you can make it pretty automatic to making copies of your repo: https://docs.superset.sh/setup-teardown-scripts

Ours are a bit complex but here's an example: https://github.com/superset-sh/superset/tree/main/.superset

Why do folks choose you guys over Conductor?

biggest thing is flexibility on what harness you use. we don't build on the claude code /codex sdk. instead, we just let you bring whatever CLI agents you prefer and just optimize the terminal experience for using CLI agents (track running / needs attention state, set up worktrees, split pane, etc)

Still misses entirely the way software is constructed. These whack-a-mole tools are pointless.

definitely understand the perspective. I feel like theres some dev work that completely requires me to be single threaded and just think hard, but there other modes where i can jump into whack-a-mole mode and try to work on a bunch of github issues in parallel, or sentry issues, or little ui bugs and the thinking goes more into reviewing plans and prs and i dont have to be as attentive to each individual session

There are individual tasks that can be “swarmed”.

Constructing software requires supervision and a continuous feedback loop from a discerning human.

agreed

Care to elaborate

Even with the strongest guardrails, no llm can be left unsupervised. And believing an llm can supervise itself is madness. For every agent you deploy as a “team” you will have multiplied diminishing returns.

I’ve iterated the process hundreds of times and even with a strong specification and tests, an llm can meet the spec and pass all the tests and still build the wrong thing.

This multi-agent workflow idea is worse than web3.

But it’s a YC thing so someone will make money.

I also don't believe in this multi-agent swarm thing. I just use Claude Code "manually" and this is still useful for that.

Why support each agent individually instead of using ACP and get much better agent coverage?

we're actually currently evaluating integrating acp for our chat, but when you start digging a bit deeper its not a complete silver bullet.

ACP is pretty much dead for Claude subscription usage

Binding the shell <-> local git clone automatically feels like the future. Great work.

thanks! give it a try :) https://github.com/superset-sh/superset

How does this compare to Cursor?

What happens if Cursor makes the exact same features as your product?

Actually Cursor is starting to converge with us as we speak! You can look at their new agents mode (which is now their default for new users) as an example.

For what happens, in our heads the end goal is building a software factory where dozens to hundreds of agents are always running - something that nobody has nailed the experience for yet. Until that's a solved problem I hope we have room to grow and build!

can you make Superset work on folders that aren't git repositories?

yes, we do support the ability to open a folder and have superset initialize it as a git project too. Would that work for your usecase?

zed , orca , /.+mux.*/ , ...

they all look incredibly / increasingly the same?

yeah i think theres a lot of ux conventions that are starting to get figured out, but we do want to be different. At least right now most dont well support remote workspace, issue tracking, or review. I bet most of the current ux patterns will look very different in a year

I like this app, but 40 year old me says I will buy it for onetime payment rather than 20 dollars monthly. Great app

Ok I used it for 2 hours and I am ok paying 50 dollars for one time. It feels so light compared to my vscode usage. I don’t know if Anthropic severs are having a good time or is it superset :)

wow they updated their software when a claude code terminal session was alive. Felt like magic. Thanks folks. please give an option to pay one time :)

No linear integration in free version and taxing it 20$/m is a bit steep.

That's fair! We do have more paid features (a slack integration, remote workspaces etc.) but yeah we haven't found the best balance for which tier to put each in for sure.

Anyways great work and piece of nice software

thanks!

It's nice to see people building things, but honestly I found the demo video a bit disappointing. A bit too slow, a bit too choppy, a bit hand wavy. It didn't make me grasp why I needed this in my life.

Windddddddoooowwwwwssssss

sooooooooooooooon! nobody on the team daily drives windows yet so until we have an engineer on the team driving windows we probably wont be able to properly support it.

I thought this was somehow related to Apache Superset.

https://superset.apache.org/

Ah actually we're unaffiliated!

Confusing name. Superset is already an established analytics tool.

yes, bad choice on my part. the origin was i was planning for it to be a superset of all your dev tools, not thinking about apache superset at all since it was a different domain

> Just realized that I have done all my work in @superset_sh since Dec 26.

Just FYI the first quote on your site references a date we haven’t reached yet!

I think Abhi meant Dec 26, 2025

How many "IDEs for the agentic era" do we need?

As many as we can new idea's from and figure out how to do this more efficiently. It's ok to stick with vim, of course!

Fair pushback! The space will settle down eventually, it's just clear to a lot of people that there's a lot of value to be created in this space that hasn't been created yet :)

We live in this era when folks can vibecode entire startups without ever making a simple Google Search.

https://github.com/apache/superset

"Please don't complain about tangential annoyances—e.g. article or website formats, name collisions, or back-button breakage. They're too common to be interesting."

https://news.ycombinator.com/newsguidelines.html

Always preferred the name Caravel...

yeah I was confused coming across this headline - "Huh? Apache Superset is doing a Launch HN?"

Launch HN: Superset (YC P26) – IDE for the agents era