I built a website that compares 68 of these tools: https://agentmgmt.dev/

Offrun is added.

All of the handful of tools in this space that I've tried have modeled an overlap/coupling between project, directory, and machine. For my workloads specifically I think these should be organized differently. Something like top-level projects that have child task/goal/sessions with each one fanning out to different agents running potentially in different worktrees, directories, and even on different machines. E.g. I have a lot of notes and other context stored locally. On a different machine I have repositories checked out and build tooling installed. Frontend and backend exist in different directories but can both be relevant to a given unit of delivery. Are there any agent orchestrators that can model things that way?

We already have similar products, like Goose [0], Paseo [1], and T3 Code [2].

[0]: https://goose.ai

[1]:https://paseo.sh/

[2]: https://t3.codes/

How's this different to https://paseo.sh/ or https://www.onorca.dev/ which seems to be more feature complete? I personally use and like Paseo. I get to connect to my LLM agents even on other machines, remotely via Tailscale and the mobile app UX is quite good.

One thing about these tools - and, as discussed, there are many - that they mostly seem to share is that they all show coding agents work as a transcript of a conversation.

Why?

The underlying conversation has nothing to do with the work. That there is one context window is an implementation detail.

I've built one of these for myself, and instead it shares questions, plans, findings, assumptions as distinct UI elements. The conversation is hidden: your focus is on "what's next".

Are there other meta-harnesses doing the same thing? I'd love to stop building my own and use a product someone else has already built!

Love these threads for sharing experience. Using Orca has been a game changer (which I found and started using because I saw somebody using Conductor)

These sorts of management tools really speed up development by automating all the PR management, such as making new worktrees, etc. and then super speed up the multitasking between different parallel feature development.

Excited to see if this one works better for me, and to try all the new alternatives in these comments.

The obvious question -- which you should spend time answering -- is what does this give that Herdr doesn't have? What's the differentiator?

I still believe this space to be unclaimed, and I am yet to see who's going to crack the demand and needs of the niches out there.

I don't think we even have a great understanding of the market, the niches, and where it is going.

It's also evolving very aggressively constantly, sticky workflows get replaced dramatically whenever new UX or tooling brings a shift into how things can be done.

I'm very curious if anyone has good research or idea of this space mapped out somewhere, I am trying to build for it too but it's really foggy right now and the North Star moves often.

It's kinda complementary to my tool for measuring ai agent session efficiency in terms of SDLC pipeline https://github.com/extractumio/todobem

Huh. I just got pitched something like this earlier this week by an agent as well.

I'm not the target market for this, but this one has the same limitation as the one that pitched me: Mac Silicon only. No Linux, no Windows, no Intel Mac either.

Sibling comment mentioned Herdr, and Herdr is on all those platforms and apparently has raised $6 Mil.

The other catch for all of these: do you integrate with my custom harness? What do you offer that I can't vibe code myself?

Thing I'd want surfaced is per account rate limit headroom before I dispatch, not after I'm stuck mid task.

Why does "and what every account has left." activate the "written by LLM" part of my brain?

I think every single one of us who use more than one agent locally has built something like this. There is definitely a need, and appreciate it being 100% free, but eventually, something needs to be so good and with so much community best practices in it, that it will beat a DIY one. E.g. I usually use Claude Opus 5.5 as the "driver" and let it spin cursor, codex, other claude subagents at will for different tasks (this way it always know what is running where). Simply asking it to break down the backlog into a DAG, show which one needs my input, which one is blocked, publish an HTML page and keep it up to date (and sinked with the built in TODO list) is usually all it takes.

I would go as far as making this a skill perhaps, so I can get the same results on other projects... but not sure what is the benefit of making it a Mac only binary, it can be a simple HTML / jsx, with maybe hooks or a /loop to keep it up to date.

A solution needs to be really better than what I can ask Claude to build in 20 minutes. It's not the "someone else spent time and thought and tokens so you don't have to" which does have some value, it's knowing that whatever I try, it will be a little different from what I want, and the energy to fork it, tweak it, is too high vs just asking Claude to build it for me on the spot...

tl;dr Definitely cool, definitely needed, but the bar for everything (including free and open source) has gone extremely high. Things that would become beloved de-facto "industry standard" tools a la "oh are you still using X everyone is using Y" just a few years ago, are now shrugged as a vibe coded lunchtime project (the new weekend project)

It's impossible to impress anyone anymore. I feel dead inside. We all got accepted to hogwarts and we became desensitized to having magic powers.

Congrats on this ship, it's hard to get anything out there!

However, as many other comments have mentioned -- how do you plan to get users in a crowded space?

how can I connect to Supero.dev via MCP to build a multi tenant app on supero.dev platofrm? any idea ?