Herdr Crash Course — Step-by-Step Setup

🧰 What Herdr Is

Herdr is a terminal multiplexer built for running AI agents. It lets you run multiple agents in parallel across workspaces, tabs, and panes — keeping them running as background processes even when you close the app or your laptop, and connecting to a VPS so agents keep working remotely. 0:00

The honest first impression: "the first time I used Herdr, I didn't like it — I was already using Cmux and Tmux. But then I tried it again two weeks ago and haven't been able to stop using it." It keeps track of all agents, keeps them running as background processes, and lets you run them in the cloud. 1:35

⚡ Install & Quick Tour

Go to herdr.dev, copy the install command, paste it into your terminal — then run herdr to open the workspace. 1:59

What it does: "it lets you run multiple applications inside a single terminal window." One Ghostty window holds many workspaces, each with multiple tabs, each tab splittable into panes. 2:20

🗂️ Workspaces, Tabs, Panes

"Ideally: one workspace per project, one tab per process or agent — an implementer here, a reviewer here, a server here." 2:40

Concept	What it is	Shortcut
Workspace	Like a window tab — one per project	`Ctrl+B` + number
Tab	One per process/agent	`Ctrl+B` + `c`
Pane	Split a tab vertically/horizontally	`Ctrl+B` + `%` / `"`
Detach	Leave the session (agents keep running)	`Ctrl+B` + `q`
Shortcuts	Show available bindings	`Ctrl+B` + `?`
Settings	Themes, notifications, labels	`Ctrl+B` + `s`

Mouse-friendly: "when you don't know the shortcuts, just click around and right-click to rename/new tab." Alejandro recommends using the mouse first, then learning key bindings one at a time. 3:10

👁️ Agent Detection & Tracking

"When you run an agent, Herdr automatically detects that this is an agent and lists it in the system bar." It shows every agent across all tabs and workspaces. 3:37

The color system: a working agent shows yellow; when it finishes it turns blue and fires a notification. "You can keep track of a bunch of agents working at the same time." 5:00

Agents survive closing the app: detach, close the terminal, "go have lunch, come back, run Herdr again — the process is still running." Your agents keep working as background processes. 5:37

🔀 What a Multiplexer Is

"A multiplexer is the layer that sits between your terminal and a bunch of processes." Each workspace/tab/pane is an individual background process — which is why closing the window doesn't kill them. If the app errors or shuts down, the sub-processes continue. 6:20

Why not Ctrl+T / Cmd+T? those belong to the terminal emulator (Ghostty), not the multiplexer. The prefix solves the ambiguity of "which app does this shortcut target?" 7:30

⌨️ Prefix Keys

"You add a prefix, and whatever comes after is a shortcut for the multiplexer. Without a prefix, it targets the app underneath." 8:20

The mental model: Ctrl+D (no prefix) goes to your coding agent and closes it. Ctrl+B (prefix), then a key, controls Herdr. This is how Tmux works too. 8:25

Settings (Ctrl+B then s): themes, indicator style, notification sounds, toast vs system notifications (off by default — enable if you want them), pane labels, and agent integrations (the natively-supported agents are listed). 8:55

☁️ Herdr on a VPS

"Run your agents remotely — turn off your computer and they keep working on the VPS, controllable from your phone." 11:10

The naive way (don't do this): SSH into the VPS, install Herdr, run it there. "It works, but adds a lot of latency because the interface runs on the server and streams back." 11:20

The right way — zero latency: herdr --remote, connect to your SSH identity. First connect asks you to install Herdr on the server (just press enter). "Herdr runs on our computer but controls the VPS — no latency. Every process runs directly in the VPS." 12:05

Reconnect: Ctrl+B q to detach → herdr session list to see open sessions (local + remote) → herdr remote VPS to rejoin where you left off. 12:55

🤖 Controlling Herdr with an Agent

"Herdr exposes an API through a CLI — so you can teach your agents to use that CLI to take control and spawn new sessions, tabs, panes." 13:20

The official skill: Herdr docs → "agent skill" file → copy it into your skills directory (or install via Pi skills). Then ask an agent: "create a new Herdr session, run Pi inside it, and ask it to say hello world" — and it does, spawning the sub-agent. 13:50

Why it matters: "now you can design your own skills and workflows to spawn sub-agents in different sessions and have them talk to each other — very powerful, a lot of potential." 15:20

🐍 Tau: Agents with Local Models

Tau is the agent Alejandro is building at Hugging Face — "a Python port of Pi, same architecture, exactly as minimalist, scores pretty much the same in benchmarks." Native support for local models (run it with Llama.cpp and pick your models). 14:15

Extensible: extensions can modify the terminal UI and add new sidebar widgets. It's a good fit if you like Pi but want a Python port or an interface closer to OpenCode. 14:35

💡 Key Takeaways

Herdr = a multiplexer for agents — run many agents in parallel across workspaces/tabs/panes, all tracked in one bar.
Agents are background processes — close the app, they keep running; detach with Ctrl+B q.
Prefix keys disambiguate shortcuts — Ctrl+B targets Herdr; no prefix targets the agent underneath.
Use herdr --remote for VPS — zero latency, unlike running the full interface over SSH; reconnect with herdr remote VPS.
Agents can control Herdr — via the CLI API + official skill, spawn sub-agents that talk to each other.
Color + notifications = situational awareness — yellow while working, blue when done, toast on completion.

📍 Timestamp Index

0:00 Intro

1:59 Install & tour

2:40 Workspaces/tabs/panes

3:37 Agent detection

5:37 Survive closing the app

6:20 What a multiplexer is

8:20 Prefix keys

11:10 Herdr on a VPS

13:20 Controlling Herdr with an agent

14:15 Tau (local models)