Herdr Crash Course — Step-by-Step Setup

2026-08-28 · Alejandro AO · tutorial · 16:43 · watch on YouTube ↗

developer-toolsterminalmultiplexerai-agentsherdervpstutorialagent-orchestration

Verdict: recommended-second-chance-convert

TL;DR

  • Alejandro AO (AI educator, builds the Tau agent at Hugging Face) delivers a structured Herder crash course: demo → what a multiplexer is → local + VPS setup → having an agent control Herder itself. Notably a second-chance conversion: "the first time I used Herder... I didn't like it. Then I tried it again two weeks ago and I haven't been able to stop using it."
  • The pedagogy is the value: the clearest explanation in the archive of the workspace→tab→pane hierarchy (one workspace per project, one tab per agent/process — implementer, reviewer, dev server) and of why multiplexers need a prefix key (Ctrl-T would hit Ghostty; unprefixed keys must reach the agent below).
  • Practical VPS detail the other reviews skipped: don't run Herder inside SSH (the UI streams back with latency) — run herder --remote <ssh-host> locally so the interface is local while every process runs on the VPS (demoed on a Hetzner box); herder session list shows local vs remote sessions.
  • The finale is the inversion demo: install Herder's official agent skill, then ask Tau to "create a new Herder session, run Pi inside of it, and ask it to say hello world" — an agent spawning and orchestrating other agents through the multiplexer's CLI/API.
  • Settings tour bonus: notification toasts are OFF by default (enable in prefix-S settings: in-Herder / terminal / system), plus themes, indicators, sounds, pane labels, and the native agent integrations list. Ongoing plug for Tau: a Python port of Pi with local-model support (llama.cpp) and UI-extending extensions.

Key moments

  • [00:01] What Herder is — run multiple agents, track them in the sidebar, background persistence, VPS-friendly (frame_0001.jpg shows the "Herdr — Terminal multiplexer for AI agents" branding).
  • [01:32] The second chance — didn't like it six weeks ago; two weeks ago "I haven't been able to stop using it."
  • [02:54] Mental model — one workspace per project; one tab per process: "an implementer here, a reviewer here, then your npm run dev."
  • [03:38] Agent auto-detection — running Pi registers it in the agents pane; working agents show yellow, done flips to blue + notification ([05:03]).
  • [06:20] Multiplexer explained — the layer between terminal and processes; why background persistence works; the prefix-key rationale ([07:36]: unprefixed keys must reach the coding agent, not the multiplexer).
  • [08:51] Settings tour — prefix-S: themes, indicators, sounds, toasts (OFF by default — enable in-Herder/terminal/system, frame_0046.jpg), pane labels, native integrations list.
  • [11:14] The VPS latency lesson — running Herder inside SSH streams the UI back ("a lot of latency you don't need"); herder --remote keeps the UI local, processes remote (Hetzner Ubuntu box, frame_0056.jpg).
  • [12:49] Session management — herder session list distinguishes the local default session from the server one; herder remote vps reattaches where you left off.
  • [13:42] The official skill — from Herder's docs; installable via npx skills or copied into your skills directory.
  • [14:57] The inversion demo — Tau: "create a new Herder session, run Pi inside of it and ask it to say hello world" — it loads the skill, creates the workspace, starts Pi, prompts it, reads the reply (frame_0076.jpg): "you can design your own workflows to spawn sub-agents... and talk to them between each other."
  • [16:09] Tau plug — Python port of Pi ("scores pretty much exactly the same"), native local models via llama.cpp, extensions that add sidebar widgets.

Hook microscope (0-10s)

  • Frames: 20 at 2 fps
  • Word-level transcript (40 words):
  [  0.38s] Good
  [  0.74s] morning
  [  1.08s] everyone
  [  1.48s] How's
  [  1.54s] it
  [  1.64s] going
  [  1.76s] today
  [  2.34s] Today
  [  2.70s] we're
  [  2.80s] going
  [  2.92s] to
  [  3.04s] be
  [  3.14s] taking
  [  3.34s] a
  [  3.50s] very
  [  3.74s] quick
  [  3.90s] look
  [  4.08s] at
  [  4.38s] this
  [  4.70s] tool
  [  4.98s] that
  [  5.12s] you
  [  5.26s] see
  [  5.40s] right
  [  5.54s] here
  [  5.90s] which
  [  5.90s] is
  [  6.34s] Herder
  [  7.24s] Now
  [  7.72s] Herder
  [  7.78s] allows
  [  8.08s] you
  [  8.24s] to
  [  8.48s] run
  [  8.62s] multiple
  [  9.10s] agents
  [  9.46s] at
  [  9.66s] the
  [  9.82s] same
  [  9.96s] time

Hook pattern: capability rundown, tutorial-style. No drama, no tease — the first 20 seconds are a feature enumeration over the product on screen ("run multiple agents at the same time and keep track of everything... workspaces... tabs... agents in the bottom"), followed immediately by the practical payoffs (background persistence, VPS control, "continue working with your agents on, say, your phone") and an explicit agenda at 0:54 (demo → multiplexer concept → local + VPS setup → agent controls Herder). The educator's hook: promise the syllabus, not the mystery — it selects for viewers who already want the skill rather than maximizing raw retention.

Editorial profile

No scene-change data — likely a static/screen-recorded source.

Screencast tutorial (~4.8 cuts/min, no talking-head-only shots): full-screen Herder/terminal capture with persistent picture-in-picture face cam, agenda shown as a cat in-this-video.md in the terminal, calm structured narration with section handoffs ("okay, let's do that") — educator format, zero B-roll.

Quotable moments

  • [01:41] "I was already using tmux, didn't really feel like Herder was adding anything to my workflow. But then I tried it again... and I haven't been able to stop using it."
  • [02:56] "Ideally, you want one workspace per project, and one tab per process — an implementer right here, a reviewer right here, then your npm run dev."
  • [11:36] "That's going to add a lot of latency that you don't need." (on running Herder inside SSH instead of --remote)
  • [15:27] "Now you can design your own skills and your own workflows to spawn up sub-agents or different sessions and talk to them between each other."

Entities mentioned

  • People: alejandro-ao (host), Ryan Dolphin (Tau-Herder extension author)
  • Companies: hugging-face (where Tau is developed), hetzner (VPS demoed)
  • Tools / products: herder, tau-agent, pi-agent, tmux, ghostty, llama-cpp, open-code (UI comparison)
  • Places: none stated

Concepts surfaced

  • agent-orchestrating-multiplexer: the inversion — an agent using Herder's CLI/API (via the official skill) to spawn sessions, run other agents inside them, and relay messages; multiplexer as sub-agent runtime.
  • workspace-layer: reconfirmed with the cleanest mental model yet: workspace = project, tab = process/agent role (implementer, reviewer, server), pane = view.
  • prefix-key-rationale: why multiplexers need Ctrl-B — key routing between three layers (terminal emulator, multiplexer, the agent/process underneath).
  • thin-client-workflow: fourth confirmation, with the technical why: --remote keeps the UI local and processes remote, versus SSH-then-herder which streams the whole interface back with latency.
  • agent-skill-file: Herder ships an official skill in its docs; installable via npx skills or copy-into-skills-dir.
  • second-chance-adoption: tools landing on retry — initial rejection ("wasn't adding anything"), later conversion once the workflow need (multiple agents) materialized.

Transcript

Source: captions.

[00:02] Good morning, everyone. How's it going today? Uh today we're going to be taking
[00:03] today? Uh today we're going to be taking a very quick look at this tool that you
[00:05] a very quick look at this tool that you see right here, which is Herder. Now,
[00:07] see right here, which is Herder. Now, Herder allows you to run multiple agents
[00:10] Herder allows you to run multiple agents at the same time and keep track of
[00:12] at the same time and keep track of everything that they are doing. It has
[00:14] everything that they are doing. It has multiple workspaces. It allows you to
[00:16] multiple workspaces. It allows you to create new tabs, and it automatically
[00:18] create new tabs, and it automatically shows the agents that are open right
[00:21] shows the agents that are open right here in the bottom, okay? Um it is also
[00:23] here in the bottom, okay? Um it is also a terminal multiplexer, which means that
[00:25] a terminal multiplexer, which means that when you close your computer or close
[00:27] when you close your computer or close the application, your agents actually
[00:29] the application, your agents actually keep running because they are background
[00:31] keep running because they are background processes, and it works great with VPSs,
[00:35] processes, and it works great with VPSs, with remote servers, so that you can
[00:37] with remote servers, so that you can connect your Herder instance to a VPS
[00:40] connect your Herder instance to a VPS and have your agents running that VPS
[00:42] and have your agents running that VPS and control them only from your
[00:43] and control them only from your computer, so you can just turn off your
[00:45] computer, so you can just turn off your computer and continue working with your
[00:47] computer and continue working with your agents on, say, your phone, for example.
[00:50] agents on, say, your phone, for example. Uh so, today we're going to be taking a
[00:52] Uh so, today we're going to be taking a look at everything about it.
[00:54] look at everything about it. First of all, we're going to see a demo,
[00:56] First of all, we're going to see a demo, then we're going to see very quickly
[00:57] then we're going to see very quickly what is a multiplexer and how it works.
[00:59] what is a multiplexer and how it works. Then we're going to see how to set this
[01:01] Then we're going to see how to set this up for local and remote work with a VPS,
[01:04] up for local and remote work with a VPS, and finally, I'm going to show you how
[01:05] and finally, I'm going to show you how to use an agent to control the Herder
[01:09] to use an agent to control the Herder API itself and have an agent create new
[01:11] API itself and have an agent create new sessions, open new tabs, etc. So,
[01:14] sessions, open new tabs, etc. So, without any further ado, let's actually
[01:16] without any further ado, let's actually get right into it.
[01:23] All right. So, uh let's actually take a look at what Herder is and how to
[01:25] look at what Herder is and how to install it and how it works. Kind of a
[01:27] install it and how it works. Kind of a very quick tour. So, uh kind of a TLDR
[01:30] very quick tour. So, uh kind of a TLDR if you want to just get right into using
[01:31] if you want to just get right into using it right away.
[01:33] it right away. Um and actually, to be completely frank,
[01:35] Um and actually, to be completely frank, the first time I used Herder was about a
[01:37] the first time I used Herder was about a month and a half ago, and I didn't like
[01:39] month and a half ago, and I didn't like it, didn't think it was for me. I was
[01:41] it, didn't think it was for me. I was already using C mux and uh T mux, so
[01:44] already using C mux and uh T mux, so didn't really feel like Herder was a-
[01:45] didn't really feel like Herder was a- adding anything to my workflow. But then
[01:48] adding anything to my workflow. But then I tried it again about a week and a half
[01:50] I tried it again about a week and a half or 2 weeks ago, and I haven't been able
[01:52] or 2 weeks ago, and I haven't been able to stop using it. It is great. It allows
[01:54] to stop using it. It is great. It allows me to keep track of all my agents and to
[01:56] me to keep track of all my agents and to keep them running as background
[01:58] keep them running as background processes, and I can also have my agents
[02:00] processes, and I can also have my agents run in the cloud, which I'm going to
[02:01] run in the cloud, which I'm going to show you how to do in just a moment. Um
[02:03] show you how to do in just a moment. Um to be
[02:05] to be uh to give you the tour, you go to
[02:07] uh to give you the tour, you go to herder.dev, you click right here on
[02:08] herder.dev, you click right here on copy, and you paste that into your
[02:11] copy, and you paste that into your terminal, and you install it. And what
[02:13] terminal, and you install it. And what that actually does is that it opens this
[02:15] that actually does is that it opens this thing right here. Let me just detach
[02:17] thing right here. Let me just detach right here.
[02:18] right here. So, you install it like this, and then
[02:19] So, you install it like this, and then you just run herder in order to open the
[02:22] you just run herder in order to open the herder workspace working at the
[02:24] herder workspace working at the application.
[02:25] application. And what it does is that it allows you
[02:27] And what it does is that it allows you to run multiple applications inside of a
[02:31] to run multiple applications inside of a single terminal window. So, as you can
[02:33] single terminal window. So, as you can see, this is still one single ghosty
[02:35] see, this is still one single ghosty terminal window, and there are a bunch
[02:37] terminal window, and there are a bunch of different workspaces right here.
[02:40] of different workspaces right here. Now, each workspace is kind of like a uh
[02:42] Now, each workspace is kind of like a uh window tab, okay? And each uh I mean,
[02:46] window tab, okay? And each uh I mean, each workspace has also multiple tabs
[02:48] each workspace has also multiple tabs that you can open within it. So, for
[02:49] that you can open within it. So, for example, here is one one workspace, and
[02:52] example, here is one one workspace, and here is two tabs that I'm running, okay?
[02:55] here is two tabs that I'm running, okay? Uh ideally, you want to have one
[02:57] Uh ideally, you want to have one workspace per project that you're
[02:58] workspace per project that you're dealing with, and one tab per process or
[03:01] dealing with, and one tab per process or different agents that are running
[03:02] different agents that are running different things. For example, I can
[03:03] different things. For example, I can have an implementer right here, a
[03:05] have an implementer right here, a reviewer right here, then a server, uh
[03:07] reviewer right here, then a server, uh your npm run dev right here, etc.
[03:10] your npm run dev right here, etc. And um
[03:12] And um also, herder is uh very mouse-friendly.
[03:14] also, herder is uh very mouse-friendly. So, when you don't know the shortcuts,
[03:16] So, when you don't know the shortcuts, you can just click around and just uh
[03:19] you can just click around and just uh right-click to rename, new tab, etc.
[03:22] right-click to rename, new tab, etc. Uh but I of course I prefer to use the
[03:24] Uh but I of course I prefer to use the keyboard, so I just do control B, and
[03:26] keyboard, so I just do control B, and for example, to create a new tab, I do
[03:28] for example, to create a new tab, I do C, and then I just name it new test, and
[03:32] C, and then I just name it new test, and then right here, I am in my new
[03:35] then right here, I am in my new uh terminal right here, and I can just
[03:36] uh terminal right here, and I can just run an agent right here, okay?
[03:38] run an agent right here, okay? And um something useful about this is
[03:40] And um something useful about this is that when you run an agent right here,
[03:42] that when you run an agent right here, for example, let's say I use pi,
[03:45] for example, let's say I use pi, it automatically detects that this is an
[03:47] it automatically detects that this is an agent, and it lists it right here in my
[03:51] agent, and it lists it right here in my system right here. As you can see, it
[03:52] system right here. As you can see, it just created a new uh system right here.
[03:54] just created a new uh system right here. I can just say hello, and here Pi Well,
[03:57] I can just say hello, and here Pi Well, I don't have
[03:58] I don't have Hi co-configured. You can just say
[04:00] Hi co-configured. You can just say hello, and here Pi responds. And this is
[04:04] hello, and here Pi responds. And this is the agent right here. You can also
[04:06] the agent right here. You can also create panes. So, for example, you can
[04:08] create panes. So, for example, you can split right or down, or you can also do
[04:11] split right or down, or you can also do control B to detach and to open a new
[04:15] control B to detach and to open a new pane right here, and you can also run
[04:16] pane right here, and you can also run multiple panes like that. Okay?
[04:18] multiple panes like that. Okay? Um so, in other words, there are
[04:21] Um so, in other words, there are workspaces that are divided into that
[04:23] workspaces that are divided into that you can divide split into tabs, and each
[04:26] you can divide split into tabs, and each tab you can split into multiple panes,
[04:29] tab you can split into multiple panes, be it vertical or horizontal like this.
[04:31] be it vertical or horizontal like this. Okay?
[04:33] Okay? And it works great. And something cool
[04:35] And it works great. And something cool about this is that when you
[04:37] about this is that when you create a new workspace or a new tab, and
[04:40] create a new workspace or a new tab, and you just open a new agent. So, for
[04:43] you just open a new agent. So, for example, here I am going to run Tau,
[04:45] example, here I am going to run Tau, which is which is the agent that I am
[04:47] which is which is the agent that I am developing right here at Hugging Face.
[04:49] developing right here at Hugging Face. You can just open Tau right here, and
[04:51] You can just open Tau right here, and you can ask it to sleep for 10 seconds.
[04:55] you can ask it to sleep for 10 seconds. And we can just go back to another
[04:57] And we can just go back to another session right here. Here are all the
[04:59] session right here. Here are all the agents across different tabs and across
[05:01] agents across different tabs and across different workspaces. And this one right
[05:03] different workspaces. And this one right here, since it has since it's working,
[05:06] here, since it has since it's working, it's shown right here in yellow. And
[05:08] it's shown right here in yellow. And when it finishes working, it will change
[05:10] when it finishes working, it will change color to blue and send me a
[05:12] color to blue and send me a notification.
[05:14] notification. Let's see what it does.
[05:16] Let's see what it does. And there you go, it's blue. And here we
[05:18] And there you go, it's blue. And here we have my notification. I can just go
[05:20] have my notification. I can just go right here and also get the nice little
[05:23] right here and also get the nice little blue thing right here on the workspace.
[05:25] blue thing right here on the workspace. So, you can open the workspace or click
[05:27] So, you can open the workspace or click right here directly in the agent, and
[05:29] right here directly in the agent, and you can see that the agent finished. So,
[05:31] you can see that the agent finished. So, you can keep track of a bunch of agents
[05:33] you can keep track of a bunch of agents working at the same time. It's very very
[05:36] working at the same time. It's very very very useful.
[05:37] very useful. And something useful again is that let's
[05:40] And something useful again is that let's for example ask it to sleep again for
[05:42] for example ask it to sleep again for other 10 seconds is you can detach from
[05:44] other 10 seconds is you can detach from it. You can close this application. You
[05:46] it. You can close this application. You can go back to go to have some food or
[05:49] can go back to go to have some food or go for lunch, then come back, run Herder
[05:51] go for lunch, then come back, run Herder again, and as you can see, the process
[05:54] again, and as you can see, the process is still running. And so, your agents
[05:56] is still running. And so, your agents will continue running in the background
[05:58] will continue running in the background even if you close the application or
[06:00] even if you close the application or leave your computer. Um and this is even
[06:03] leave your computer. Um and this is even more useful if you connect it to a VPS,
[06:05] more useful if you connect it to a VPS, which is what I'm going to show you in a
[06:06] which is what I'm going to show you in a moment how to set it up, okay?
[06:09] moment how to set it up, okay? So, that is essentially the very quick
[06:11] So, that is essentially the very quick overview of Herder and how it works.
[06:14] overview of Herder and how it works. Let's now see very quickly what is a
[06:17] Let's now see very quickly what is a multiplexer and how to set this up.
[06:20] multiplexer and how to set this up. All right. So, very quickly, let's talk
[06:22] All right. So, very quickly, let's talk about what is a multiplexer. I don't
[06:24] about what is a multiplexer. I don't want to go very much into detail with
[06:26] want to go very much into detail with it. I just want to give you exactly what
[06:27] it. I just want to give you exactly what you need to understand what's actually
[06:29] you need to understand what's actually going on and how to use Herder. So, a
[06:31] going on and how to use Herder. So, a multiplexer essentially is this layer or
[06:34] multiplexer essentially is this layer or this application that sits in between
[06:37] this application that sits in between your terminal and a bunch of different
[06:39] your terminal and a bunch of different processes, okay? So, as you saw right
[06:41] processes, okay? So, as you saw right here, this is a single Ghosty tab, a
[06:44] here, this is a single Ghosty tab, a single Ghosty window. And when I open
[06:47] single Ghosty window. And when I open Herder, there are a bunch of different
[06:49] Herder, there are a bunch of different things happening inside this individual
[06:51] things happening inside this individual window. So, I have a bunch of different
[06:53] window. So, I have a bunch of different workspaces, a bunch of different tabs, a
[06:55] workspaces, a bunch of different tabs, a bunch of different agents, and each one
[06:58] bunch of different agents, and each one of this is an individual process that is
[07:02] of this is an individual process that is running in the background. And that is
[07:04] running in the background. And that is why I can then close this window and
[07:07] why I can then close this window and it's going to continue running those
[07:08] it's going to continue running those processes. So, the multiplexer is this
[07:10] processes. So, the multiplexer is this layer in between that that handles and
[07:13] layer in between that that handles and manages all these sub-processes so they
[07:15] manages all these sub-processes so they can run in the background. And if
[07:17] can run in the background. And if there's ever an error or some or I shut
[07:20] there's ever an error or some or I shut down the application or something like
[07:22] down the application or something like that, these sub-processes are going to
[07:24] that, these sub-processes are going to continue running, and that is the good
[07:26] continue running, and that is the good part about this whole thing.
[07:28] part about this whole thing. Now, before I told you, for example,
[07:30] Now, before I told you, for example, that in order to open a new tab, you do
[07:32] that in order to open a new tab, you do control B, and then you do C, and then
[07:34] control B, and then you do C, and then just name it whatever you want, okay?
[07:37] just name it whatever you want, okay? Um now, you may be wondering, why don't
[07:39] Um now, you may be wondering, why don't I just do control T or command T?
[07:43] I just do control T or command T? Well, the reason for that is that it is
[07:45] Well, the reason for that is that it is not very straightforward to access the
[07:47] not very straightforward to access the multiplexer with shortcuts. Uh, this
[07:50] multiplexer with shortcuts. Uh, this happens of course by the way with tmux
[07:52] happens of course by the way with tmux as well. If you want to do a new tab and
[07:55] as well. If you want to do a new tab and you want it to do con- command T, that's
[07:57] you want it to do con- command T, that's going to open a new tab in Ghost T. If
[08:00] going to open a new tab in Ghost T. If you wanted to do for example some uh,
[08:03] you wanted to do for example some uh, shortcut in your uh, coding agent, how
[08:07] shortcut in your uh, coding agent, how do you know that control D is going to
[08:09] do you know that control D is going to go into your coding agent and not into
[08:12] go into your coding agent and not into your multiplexer? Well, this is the
[08:14] your multiplexer? Well, this is the reason why you need a prefix, okay? And
[08:17] reason why you need a prefix, okay? And that is how multiplexers usually work.
[08:20] that is how multiplexers usually work. You add a prefix and whatever comes
[08:22] You add a prefix and whatever comes after the prefix is a shortcut for your
[08:24] after the prefix is a shortcut for your multiplexer. And if you don't use a
[08:26] multiplexer. And if you don't use a prefix, then it targets the application
[08:28] prefix, then it targets the application or the processes that it's running
[08:30] or the processes that it's running underneath the uh, multiplexer, okay?
[08:33] underneath the uh, multiplexer, okay? So, if right here I do control D, then
[08:36] So, if right here I do control D, then that is going to go into my coding agent
[08:39] that is going to go into my coding agent and it go- it's going to close my coding
[08:41] and it go- it's going to close my coding agent. If I do control B,
[08:43] agent. If I do control B, I enable the prefix and whatever comes
[08:46] I enable the prefix and whatever comes after is going to control Herder, okay?
[08:49] after is going to control Herder, okay? So, in this case, I can just do uh,
[08:51] So, in this case, I can just do uh, question mark to open the shortcuts
[08:54] question mark to open the shortcuts available. And here I can see for
[08:56] available. And here I can see for example that prefix plus S uh, opens the
[08:59] example that prefix plus S uh, opens the settings. So, I can just do control B
[09:01] settings. So, I can just do control B again, which is my prefix, and then S to
[09:04] again, which is my prefix, and then S to open my settings. And as you can see
[09:06] open my settings. And as you can see right here, I have my themes that I can
[09:08] right here, I have my themes that I can run. I have my indicators, so for
[09:10] run. I have my indicators, so for example, here I can use these or this
[09:13] example, here I can use these or this one right here. The sound, if I want my
[09:15] one right here. The sound, if I want my notifications to enable any sound, uh,
[09:18] notifications to enable any sound, uh, here I have uh, if I want my
[09:20] here I have uh, if I want my notifications to open toast right here
[09:22] notifications to open toast right here on the on the um, in the corner of
[09:24] on the on the um, in the corner of Herder, or if I want them to be
[09:26] Herder, or if I want them to be notifications of the terminal or
[09:27] notifications of the terminal or notifications via system, I can set it
[09:30] notifications via system, I can set it up right here. I think by default it is
[09:32] up right here. I think by default it is off, so if you want notifications, um,
[09:34] off, so if you want notifications, um, do enable this right here.
[09:36] do enable this right here. For pane labels, if you want to label
[09:38] For pane labels, if you want to label the panes to have different labels. For
[09:40] the panes to have different labels. For integrations, in this case these are all
[09:42] integrations, in this case these are all the agents that are natively supported.
[09:45] the agents that are natively supported. And um yeah, that's basically how it
[09:47] And um yeah, that's basically how it works. If you want to use Tao, for
[09:50] works. If you want to use Tao, for example, what you're going to want to do
[09:52] example, what you're going to want to do is you're going to want to install the
[09:54] is you're going to want to install the Tao extension. So, Tao install. Then
[09:56] Tao extension. So, Tao install. Then you're going to want to use Tao Herder
[09:59] you're going to want to use Tao Herder right here by Ryan Dolphin. And then
[10:01] right here by Ryan Dolphin. And then you're going to be able to use uh Tao
[10:04] you're going to be able to use uh Tao within
[10:06] within uh Herder. Okay?
[10:07] uh Herder. Okay? And uh there you go. Essentially, that
[10:10] And uh there you go. Essentially, that is what is going on right here. And
[10:12] is what is going on right here. And that's what is a multiplexer. I do
[10:14] that's what is a multiplexer. I do encourage you to use the mouse at first
[10:16] encourage you to use the mouse at first and to identify what you're uh more
[10:18] and to identify what you're uh more comfortable with doing. And then
[10:20] comfortable with doing. And then whenever you want to do something, just
[10:22] whenever you want to do something, just take a look at the key bindings and
[10:25] take a look at the key bindings and slowly get used to these shortcuts so
[10:28] slowly get used to these shortcuts so that you can actually start using Herder
[10:30] that you can actually start using Herder without using the mouse because of
[10:31] without using the mouse because of course it's nicer to just use the
[10:33] course it's nicer to just use the keyboard. Okay?
[10:35] keyboard. Okay? Now, in order to detach from a Herder
[10:37] Now, in order to detach from a Herder session that is currently open, all you
[10:40] session that is currently open, all you want to do is control B and then Q.
[10:42] want to do is control B and then Q. And now your all your agents will
[10:45] And now your all your agents will continue running and everything that you
[10:46] continue running and everything that you started right there, your processes,
[10:48] started right there, your processes, your servers, everything's going to
[10:50] your servers, everything's going to continue running. And then when you go
[10:52] continue running. And then when you go back to Herder, it's going to be right
[10:54] back to Herder, it's going to be right where you left off. Okay?
[10:56] where you left off. Okay? Um so that is essentially how to set up
[10:58] Um so that is essentially how to set up Herder uh for your local machine right
[11:00] Herder uh for your local machine right here. Everything is local. Uh let's take
[11:02] here. Everything is local. Uh let's take a look at how to do the same thing but
[11:05] a look at how to do the same thing but on a VPS so that you can even turn off
[11:07] on a VPS so that you can even turn off your computer and your agents keep
[11:09] your computer and your agents keep running on the VPS and you can control
[11:11] running on the VPS and you can control them with Herder. Okay? Let's do that.
[11:14] them with Herder. Okay? Let's do that. All right. So, let's take a look at how
[11:16] All right. So, let's take a look at how to run your agents remotely in your VPS.
[11:18] to run your agents remotely in your VPS. Uh it's very straightforward, actually.
[11:20] Uh it's very straightforward, actually. Uh well, of course the first uh thing
[11:22] Uh well, of course the first uh thing that you can do is to just connect to
[11:24] that you can do is to just connect to your VPS via uh SSH. And then inside of
[11:27] your VPS via uh SSH. And then inside of here, install Herder like we did before
[11:30] here, install Herder like we did before and then just run Herder
[11:31] and then just run Herder like that. And that's going to work.
[11:34] like that. And that's going to work. That's going to open the Herder
[11:35] That's going to open the Herder interface in your VPS, but it's going to
[11:37] interface in your VPS, but it's going to add a lot of latency because Herder the
[11:39] add a lot of latency because Herder the Herder interface is going to be running
[11:41] Herder interface is going to be running in your
[11:42] in your server and then streaming that interface
[11:45] server and then streaming that interface back into your terminal. So, that's
[11:46] back into your terminal. So, that's going to add a lot of latency that you
[11:49] going to add a lot of latency that you don't need.
[11:50] don't need. What we're going to do instead
[11:52] What we're going to do instead is we're going to disconnect from it and
[11:54] is we're going to disconnect from it and we're going to run Herder like this,
[11:57] we're going to run Herder like this, dash dash remote, and we're going to
[11:59] dash dash remote, and we're going to connect to the same SSH identity that we
[12:03] connect to the same SSH identity that we have right here, connect. If this is the
[12:05] have right here, connect. If this is the first time that you're connecting, then
[12:06] first time that you're connecting, then it's going to ask you to
[12:09] it's going to ask you to install Herder first and you just click
[12:11] install Herder first and you just click enter to validate it. It's going to
[12:12] enter to validate it. It's going to install Herder and right here we are
[12:14] install Herder and right here we are running Herder within our computer, but
[12:17] running Herder within our computer, but it is controlling our VPS. So, as you
[12:19] it is controlling our VPS. So, as you can see, there is no latency.
[12:21] can see, there is no latency. Everything's running fine. But once I
[12:24] Everything's running fine. But once I click enter, every sub process or every
[12:26] click enter, every sub process or every process is going to be run directly in
[12:28] process is going to be run directly in the VPS. So, I can just run this right
[12:30] the VPS. So, I can just run this right here, turn off my computer, and that's
[12:33] here, turn off my computer, and that's going to keep working on my VPS. Okay?
[12:36] going to keep working on my VPS. Okay? So, right here I can just open my
[12:39] So, right here I can just open my agents. I can just create and install my
[12:41] agents. I can just create and install my GitHub credentials, my Git credentials,
[12:44] GitHub credentials, my Git credentials, clone my repositories, install my
[12:46] clone my repositories, install my agents, have them all work right here,
[12:48] agents, have them all work right here, and then detach from it. I can just do
[12:50] and then detach from it. I can just do control B Q to detach. Then I can just
[12:53] control B Q to detach. Then I can just do Herder session list to see which
[12:55] do Herder session list to see which sessions are open and as you can see, it
[12:57] sessions are open and as you can see, it is my default session right here, which
[12:59] is my default session right here, which is open in this machine and the other
[13:02] is open in this machine and the other one is in the server. So, in order to
[13:04] one is in the server. So, in order to reconnect to it, I just do Herder remote
[13:06] reconnect to it, I just do Herder remote VPS and I am back where I left off, as
[13:09] VPS and I am back where I left off, as you can see. Okay? So, there you go. Not
[13:12] you can see. Okay? So, there you go. Not going to go much further about this.
[13:14] going to go much further about this. Let's take a look at how to have your
[13:15] Let's take a look at how to have your agents control your Herder session.
[13:18] agents control your Herder session. All right. So, very quickly I'm just
[13:20] All right. So, very quickly I'm just going to show you how to have your
[13:21] going to show you how to have your agents control Herder. Now, Herder it's
[13:25] agents control Herder. Now, Herder it's actually very straightforward because
[13:27] actually very straightforward because Herder exposes an API through SCLI. So,
[13:31] Herder exposes an API through SCLI. So, you can just teach your agents to use
[13:32] you can just teach your agents to use that CLI in order to take control over
[13:35] that CLI in order to take control over Herder and have them initialize new
[13:37] Herder and have them initialize new agents, new workspaces, new tabs, new
[13:40] agents, new workspaces, new tabs, new panes, etc. So,
[13:42] panes, etc. So, there is no better way to do this than
[13:44] there is no better way to do this than with a skill and they have an official
[13:47] with a skill and they have an official skill actually.
[13:48] skill actually. So, the best way to find it is to go
[13:50] So, the best way to find it is to go right here to Herder documentation. You
[13:52] right here to Herder documentation. You go to the agent skill file and right
[13:54] go to the agent skill file and right here you will find it. Here it is. And
[13:58] here you will find it. Here it is. And as you can of course install it with the
[14:00] as you can of course install it with the skills within PX skills or you can just
[14:03] skills within PX skills or you can just copy this directly into your skills
[14:04] copy this directly into your skills direct directory and then once that is
[14:07] direct directory and then once that is done, you will have it right here. I do
[14:10] done, you will have it right here. I do encourage you to use Tau. It is the
[14:12] encourage you to use Tau. It is the agent that we've been building right
[14:13] agent that we've been building right here. It is really fun. It comes with
[14:16] here. It is really fun. It comes with native support for local models. So, you
[14:18] native support for local models. So, you can just go right here and run it with
[14:20] can just go right here and run it with llama CPP and select your models and it
[14:23] llama CPP and select your models and it is actually built on top of Pi or well,
[14:26] is actually built on top of Pi or well, more like
[14:27] more like using the exact same architecture of Pi.
[14:29] using the exact same architecture of Pi. You can think about it as a port a
[14:31] You can think about it as a port a Python port of Pi. It is exactly as
[14:34] Python port of Pi. It is exactly as minimalist and it scores pretty much
[14:35] minimalist and it scores pretty much exactly the same way in most benchmarks
[14:38] exactly the same way in most benchmarks because it is pretty much the same thing
[14:40] because it is pretty much the same thing but in Python. You can also extend it
[14:42] but in Python. You can also extend it with extensions and the extensions can
[14:44] with extensions and the extensions can also modify the terminal user interface
[14:46] also modify the terminal user interface and add new widgets right here to your
[14:48] and add new widgets right here to your sidebar. So, that is pretty cool.
[14:50] sidebar. So, that is pretty cool. And as you can see right here, I already
[14:51] And as you can see right here, I already have loaded Herder as you can see. And
[14:54] have loaded Herder as you can see. And so, let me just start a new session
[14:57] so, let me just start a new session right here.
[14:58] right here. Open Tau and ask it
[15:01] Open Tau and ask it create a new Herder session
[15:04] create a new Herder session run Pi inside of it and ask it to say
[15:06] run Pi inside of it and ask it to say hello world.
[15:08] hello world. And let's see. Just like in Pi, we can
[15:11] And let's see. Just like in Pi, we can just enable to see the thinking traces
[15:13] just enable to see the thinking traces with control T.
[15:22] And there you go. As you can see, we have
[15:23] have here Pi agent saying hello world and as
[15:25] here Pi agent saying hello world and as you can see, this can be very powerful
[15:28] you can see, this can be very powerful because now you can
[15:29] because now you can design your own skills and your own
[15:31] design your own skills and your own workflows to
[15:34] workflows to spawn up sub agents or different
[15:36] spawn up sub agents or different sessions
[15:37] sessions and talk to them between each other with
[15:39] and talk to them between each other with this Herder skill, which is pretty
[15:41] this Herder skill, which is pretty pretty fun. I do encourage you to
[15:43] pretty fun. I do encourage you to experiment with it. I have not finished
[15:44] experiment with it. I have not finished experimenting with this particular part,
[15:46] experimenting with this particular part, but it looks really really fun and with
[15:48] but it looks really really fun and with a lot of potential.
[15:50] a lot of potential. So, definitely let me know what you
[15:52] So, definitely let me know what you recommend that I try about this.
[15:55] recommend that I try about this. So, there you go. That is essentially
[15:56] So, there you go. That is essentially everything that there is to know to
[15:59] everything that there is to know to kickstart with Herder. I have been very
[16:01] kickstart with Herder. I have been very much enjoying it. So, big shout out to
[16:05] much enjoying it. So, big shout out to the creators. Thanks a lot to you for
[16:07] the creators. Thanks a lot to you for watching. Do give a try to Tau. I've
[16:10] watching. Do give a try to Tau. I've been putting a lot of effort to it,
[16:12] been putting a lot of effort to it, but especially if if you're familiar
[16:15] but especially if if you're familiar with Pi or you really like Pi and you
[16:16] with Pi or you really like Pi and you want a Python port of it or you want
[16:20] want a Python port of it or you want a user interface that looks a little bit
[16:22] a user interface that looks a little bit more like like open code.
[16:24] more like like open code. And there you go. Well, thanks a lot for
[16:25] And there you go. Well, thanks a lot for watching. It's been a great pleasure as
[16:27] watching. It's been a great pleasure as usual. Do definitely give a try to
[16:30] usual. Do definitely give a try to Herder. I completely recommend it and it
[16:32] Herder. I completely recommend it and it works extremely well with Tau, so I do
[16:35] works extremely well with Tau, so I do recommend that you give a try to that
[16:36] recommend that you give a try to that one, too.
[16:37] one, too. Thanks a lot. It has been an absolute
[16:39] Thanks a lot. It has been an absolute pleasure and I will see you in the next
[16:41] pleasure and I will see you in the next one.

All frames

Total: 80. Hero frames flagged with star.

  • frames/frame_0001.jpg (t=00:00) frames/frame_0002.jpg (t=00:13) frames/frame_0003.jpg (t=00:25) frames/frame_0004.jpg (t=00:38) frames/frame_0005.jpg (t=00:50) frames/frame_0006.jpg (t=01:03) frames/frame_0007.jpg (t=01:15) frames/frame_0008.jpg (t=01:28) frames/frame_0009.jpg (t=01:40) frames/frame_0010.jpg (t=01:53) frames/frame_0011.jpg (t=02:05) frames/frame_0012.jpg (t=02:18) frames/frame_0013.jpg (t=02:30) frames/frame_0014.jpg (t=02:43) frames/frame_0015.jpg (t=02:56) frames/frame_0016.jpg (t=03:08)
  • frames/frame_0017.jpg (t=03:21) frames/frame_0018.jpg (t=03:33) frames/frame_0019.jpg (t=03:46) frames/frame_0020.jpg (t=03:58) frames/frame_0021.jpg (t=04:11) frames/frame_0022.jpg (t=04:23) frames/frame_0023.jpg (t=04:36) frames/frame_0024.jpg (t=04:48) frames/frame_0025.jpg (t=05:01) frames/frame_0026.jpg (t=05:14) frames/frame_0027.jpg (t=05:26) frames/frame_0028.jpg (t=05:39) frames/frame_0029.jpg (t=05:51) frames/frame_0030.jpg (t=06:04) frames/frame_0031.jpg (t=06:16) frames/frame_0032.jpg (t=06:29)
  • frames/frame_0033.jpg (t=06:41) frames/frame_0034.jpg (t=06:54) frames/frame_0035.jpg (t=07:06) frames/frame_0036.jpg (t=07:19) frames/frame_0037.jpg (t=07:31) frames/frame_0038.jpg (t=07:44) frames/frame_0039.jpg (t=07:57) frames/frame_0040.jpg (t=08:09) frames/frame_0041.jpg (t=08:22) frames/frame_0042.jpg (t=08:34) frames/frame_0043.jpg (t=08:47) frames/frame_0044.jpg (t=08:59) frames/frame_0045.jpg (t=09:12) frames/frame_0046.jpg (t=09:24) frames/frame_0047.jpg (t=09:37) frames/frame_0048.jpg (t=09:49)
  • frames/frame_0049.jpg (t=10:02) frames/frame_0050.jpg (t=10:15) frames/frame_0051.jpg (t=10:27) frames/frame_0052.jpg (t=10:40) frames/frame_0053.jpg (t=10:52) frames/frame_0054.jpg (t=11:05) frames/frame_0055.jpg (t=11:17) frames/frame_0056.jpg (t=11:30) frames/frame_0057.jpg (t=11:42) frames/frame_0058.jpg (t=11:55) frames/frame_0059.jpg (t=12:07) frames/frame_0060.jpg (t=12:20) frames/frame_0061.jpg (t=12:32) frames/frame_0062.jpg (t=12:45) frames/frame_0063.jpg (t=12:58) frames/frame_0064.jpg (t=13:10)
  • frames/frame_0065.jpg (t=13:23) frames/frame_0066.jpg (t=13:35) frames/frame_0067.jpg (t=13:48) frames/frame_0068.jpg (t=14:00) frames/frame_0069.jpg (t=14:13) frames/frame_0070.jpg (t=14:25) frames/frame_0071.jpg (t=14:38) frames/frame_0072.jpg (t=14:50) frames/frame_0073.jpg (t=15:03) frames/frame_0074.jpg (t=15:15) frames/frame_0075.jpg (t=15:28) frames/frame_0076.jpg (t=15:41) frames/frame_0077.jpg (t=15:53) frames/frame_0078.jpg (t=16:06) frames/frame_0079.jpg (t=16:18) frames/frame_0080.jpg (t=16:31)