Bro vs Herdr.

The short version.

Pick Herdr for the lightest, most open way to herd any agent, in a terminal or now a native app. Pick Bro for the whole workstation: managed logins, any model in any agent, an orchestrator that keeps a schedule, and a phone to carry it all.

What each one is.

Herdr's desktop app and Bro's both landed in the last ten days, and both run real terminals drawn on the GPU. Herdr has older products underneath, so it's worth being clear which Herdr we mean.

Row by row.

The same questions for both tools. A filled dot is a yes, a half dot is partly, a dashed one is a no. Because Herdr is two programs, a cell says which one it means when they differ. Flip the switch to hide every row where you'd get the same answer from either.

Managed login. Sign in once.

Run three agents and you're quietly managing three sign-ins, often two accounts each: work and personal, Claude and ChatGPT, plus whatever Gemini is doing this week. Most tools leave that to you, and you find out a login ran dry when an agent stops halfway through a task.

Bro takes it over. Add a login from the launcher and Bro runs the sign-in, keeps it in a profile of its own and flags one that needs attention before a session finds out. After that any session can run on any login, the sidebar shows what's left on each, and in proxy mode a login at its limit hands its work to the next one before the agent has seen an error.

How Herdr handles it

It doesn't, on purpose. There's no Herdr account, no profiles for several logins and no usage meters. Each agent signs in on its own inside its terminal, and a second Claude account means a custom launch command with its own environment. Its maker describes the stance in one line, and it's a consistent one:

The upside is that Herdr never touches your credentials. The cost is that logins and limits stay your chore: its welcome screen suggests asking your own agent to build a cost meter as a mod. Several logins are your call either way. Use accounts you own, and read each provider's terms first.

Many projects at once.

Run enough agents and the problem stops being any one of them. It's six projects with two agents each, and one that has quietly been waiting on you for an hour.

Herdr handles this well. Its app organises work into spaces (the TUI calls them workspaces) with tabs and panes, and an agent's state rolls up to its space and its tab. The Agents list puts waiting agents first, lets you pin, can group by space, nests worktrees under their repo, and shows spaces on other machines in the same sidebar.

Bro groups sessions by project too, the git root or the folder when there's no repo, and rolls the state up to each project's row. Alt A jumps to whoever needs you. Where it goes further is coordination. Each project can have its own orchestrator, and a root orchestrator above them opens or reuses those and takes your voice, your phone and your schedule: say "open a session in justgains" and it finds the project. By default Claude sessions can only find and message sessions in their own project. Claude Code on its own lets any session on a machine reach any other, and nothing in Herdr's docs changes that.

A terminal UI and a desktop app that agree.

Herdr has both a terminal app and a desktop app, which is more than most tools. They're separate programs, though. A terminal you start in one doesn't show up in the other, and after a reboot the TUI can resume supported agents' conversations while the app brings back fresh shells in their last folders.

Bro treats the window as a view. The desktop app and bro-cli are one workspace, with the same sessions, sidebar, launcher, views and keys, and the sessions live in a server of their own, so closing Bro, updating it or crashing it doesn't end them. Open the app at your desk, SSH in from a train, add bro --view for another window, and each shows the same agents. After a restart Claude and Codex resume their conversations where they left off. Only a few things need a window, such as focus mode and the media panel.

Keep the agent you like. Pick the model.

Herdr has no model layer. Whatever model an agent uses is set by that agent's own command line, and API keys pass through as ordinary environment variables. That's tidy, and it also means Claude Code stays on Claude and Codex stays on OpenAI.

Bro puts a translating proxy in the middle. It speaks the Anthropic Messages, OpenAI Chat, OpenAI Responses and Gemini APIs, so Claude Code can run on GPT, Gemini, Kimi, DeepSeek or Grok, and Codex, Pi and omp can run on Claude, with tools, images and thinking intact. The catalogue holds 16 providers, OpenRouter adds 463 live models, and Ollama and LM Studio cover the ones on your own machine. You choose the model per session, not per install.

A crew, or a toolbox.

Herdr is generous with the primitives. In the app and core, agents can start other agents, send them work, wait on them and hear back, through hc run --bg, hc send, hc wait and hc events. A child reports into its parent's input, roles live in Markdown templates, and the whole surface is a CLI, an open RPC API and a skill your agent can read. There's no MCP server, though, and nothing in its docs about scheduled prompts, watchdogs or a task queue.

Bro ships the opinionated version. Its orchestrator is a session that runs the others: it opens sessions, hands out tasks, waits and tells you when they're done, on a schedule (in 10m, daily 09:00) or by voice (hold Right Win, say "open a session in justgains"). 26 MCP tools back it, callable from any shell as bro ctl. Watchdogs close the loop: when an agent goes idle, a fresh reviewer reads the chat and the code, checks your milestones and sends the next task.

On your phone, two ways.

Herdr's answer is no app at all. Its TUI adapts to narrow screens, so any SSH client works, and its docs show it running over Mosh on an iPhone. You're inside the real terminal UI, so you can read, switch spaces and reply. There's no push, and the new desktop app has no phone or web client.

Bro's phone app pairs with Bro on your computer over your own network or Tailscale and lists every session like a chat: a spinner while it works, an alert when it needs you, a check when it's done. Questions arrive as buttons, dictation runs on the phone, and tapping a file path opens the file at the line. It isn't on the stores yet, so today Herdr's no-app route is the one you can use tonight.

Batteries included, or a blank canvas.

Herdr hands you a clean frame and expects your agent to build the rest. Everything in the app is a plain file in ~/.config/herdr-app: layouts, keys, themes and mods written in JavaScript, with the line "No plugin system, no fork to maintain. You ask, it edits." The TUI has a plugin marketplace of nearly two thousand listings. If your agent can write it, Herdr can host it.

Bro ships a finished product instead. 57 themes, from VS Code favourites to skinned ones with scanlines, 19 terminal fonts, 376 sound packs that say "Work, work" when an agent finishes, 128 keybindings, a media panel that plays what your agents make, a Skills view that installs across agents and Quips for reusable instructions. There's no plugin API, so the trade is real: you get more on day one, and Herdr gets you further on day thirty if you like building your own.

Where Herdr is ahead.

A lot, and it's worth reading before you decide.

• Scale and staying power. The TUI has over 46,000 stars, 98 releases in seven months and about 1.9 million installs. The team has raised a $6M seed and is part of Y Combinator's F26 batch. Bro is far younger and smaller.
• More agents out of the box. The app's launch menu lists 15 and the TUI's docs name 23, including Cursor, OpenCode, Gemini CLI, Amp, Kimi and Kiro. Bro has four first-class agents and a plain shell for the rest.
• Agent-to-agent primitives. Parent and child agents, --outlive-parent, events, templates and an open API, plus an open status protocol (OSC 7501, written by Mitchell Hashimoto) that Claude Code, Pi and omp shipped support for this month, with Codex's change merged too.
• Git inside the window. A worktree mod, a Changes panel with per-file diffs and a commit box, and worktree commands in the TUI. Bro leaves git to your agents and your tools.
• A UI your agent can rewrite, plus about 2,000 TUI plugins and a sharing index of community layouts and mods.
• Lightness. The TUI runs wherever your terminal does, and it can resume a supported agent's conversation after a reboot.

One honest caution about the app: it's a week old, and its tracker shows it. There are no OS notifications yet, no terminal search, Macs must be Apple silicon, and an agent that doesn't report its status, such as Cursor's CLI, won't appear in its Agents list. Bro's desktop app is nine days old, so give both room to settle.

Which one, then?

Match the line that sounds like your week.

They aren't mutually exclusive. Both run the agent CLIs you already have installed, so trying one doesn't cost you the other.

Questions people ask.