threecheese
Heeler has a PR open to target the new core, but no sign anyone is working on a tui for it :(
They both follow the same "gui with a cli to manage the engine which owns the PTYs, has some interaction protocol, wraps coding agent interfaces, and you can connect to remote machines" model.
Unfortunately they are both GUI only atm and so I'm still using herdr-tui over ssh until it breaks or we get a tui surface.
I love the current model, where I can use the tui in ghosttty or wezterm or iterm, or pop an ssh session, and if it makes sense there's a gui version for whatever device I'm using. It's like magic, I'm free to do whatever from wherever. Forcing a GUI feels like them asserting ownership or control. Now I need their mobile app if i want to develop remotely, etc. I recognize the irony, the core is open- I'm free to develop against it - but we all know that realistically only a handful will and those will lag. Also i am sure that GUI is first because that's where the users are. Eh.
- it uses a very fast model for normal responses, but also does reasoning and all the agent things. It responds within a second or two. - it feels like it has multiple threads internally- it can respond to your messages and questions immediately, but also do thinking and web search like a normal thread. - it can communicate with other agent threads, as an orchestrator (although this isn't unique to it). - it has dedicated persistent compute already, with a desktop that you control, and so gives you a UI surface area that the agent can use.
And so the value for me is an always-on agent conversation, where you can tell it to perform tasks and ask it questions, and it just does all of it immediately. Those tasks can be software development instead of like ordering dinner.
It has SMS / phone comm channels also, but I don't use that.
If it can communicate with the agent, it can be used to drive the agent - to send it tasks, and communicate the status of those tasks.
And so Herdr - and Superlogical, and others, is a tool which leverages PTY ownership and terminal protocols to allow orchestration of coding agents across vendors using their CLI surfaces.
Either:
1. the training data contained samples demonstrating how to do this, or it was post trained into the model intentionally; 2. the training data contained enough predicate knowledge that they could reason through it themselves, or; 3. they invented a novel approach using their vast intelligence (/s).
These are in order of likelihood. Given the high probability that it was (1), can't we just ... not do that? Does the frontier model need to be as good at recipes as offensive cybersecurity?
1. is something that many intensely don't want, and many with valid reasons; 2. will inarguably consume some of a resource which we all need, which is already becoming more scarce globally; 3. is in many cases contributing negatively to actual or perceived quality of life via DC noise and pollution; 4. threatens the livelihood of millions in the near term, even if it will improve outcomes over time; 5. is inextricably tied to and will substantially benefit a wealthy cohort, who is already seen as exerting undue control over populations governments ; 6. present potential systemic societal risk, yet is moving forward at breakneck speed; 7. has apparent total US government support, despite complaints about all the above, in a political environment which is frankly frightening, and where that government is exercising little restraint against - and collecting more sensitive data than ever about - citizens who protest against policy.
None of us should be surprised about this reaction.
We can all throw questions like this at our LLMs, and the objections you'll see here revolve around having our time wasted by folks who believe their LLM output is an accomplishment; we're getting sick of it.
Spec docs - at least in all the other things like this thing - aren't a living representation of truth, instead they are a collection of snapshots in time of feature deliverables. Layer 500 of these on top of each other and they quickly become not just useless but actively harmful. Archive them, compile them, whatever your plugin does, it still doesn't help. Nobody is going to pay for tokens just to keep the specs time-coherent, when the agent can just read the code.
Don't get me wrong, a spec is a useful bit of kit for development, just like Jira and Rally, but also like Jira and Rally they aren't the source of truth because there's already a better one which you just paid for.
There isn't one adult who looks at this massive centralization and sees that it goes against the very principle of the thing?
And I'm fairly sure Anthropic did implement that though, at least for Claude SDK (which is/was a wrapper around claude -p); I was using the sdk for a product, and had a demo to show/ensure it was billed via subscription. I basically unset any API env vars, ensured the cli credential was in place, and spawned agents and watched the logs. When I re-ran it during that time period, it clearly went to extra data (there are messages in response logs calling it out).
Recently did a quick experiment using four python cli libraries with codex, fairly big cli surface, well-typed code with docstrings, and the differences were negligible- it came down to those two qualities specifically, and the means to those ends didn't really matter. Obviously anecdata though.
I dont even know what should be private anymore - if US federal agencies can just buy ad data, join it with Flock, sprinkle Instagram in there, and inner join Social Security dump from doge, do I just give up and put the sex toys in the parlor window? Am I one OpenAI-sued-by-a-newspaper away from giving up the ghost?
They revoked all but one developers, which is probably the right approach, but I'd pin LiteLLM to "something before this version" and keep it there until further notice!
We are giving an unprecedented amount of personal data to these tools at the same time that the technology is being leveraged against people at an unprecedented scope and scale. No US legal framework exists to protect you against this. You should take this risk seriously.
Forgive the anthropomorphizing, it was necessary - "terms" has double meaning there, and each of us can understand that without an ontology - because we can access nuance. I can read schema, structure, and shape, and understand; but only one of those triggered some readers.
Anyway, the machines will choose terms which are most well-represented in the training data. This will normalize communication by chopping off the top-end where nuance lies. This is token-efficient.
Ontologies, in their implementation in the Linked Data ecosystem, do offer the ability to relate one understanding-system to another, but this is not token-efficient. I would love to see each domain, each business even, define their view of the world in the most expressive and culture matching owl representation, but I don't see it happening. "Shape" it is.
Or, is this "people watching broadcast content on devices", or "you'll get YouTube TV for free"?
Of course there are great reasons for using pointers also. I'd say this is mostly a preference, I'm sure you are extremely effective using your own methods. I have maybe two dozen terminals open atm, and I can tell you what each of them is for - but put a phone bill or a bank statement in front of me and it might as well be Greek.