The editorial argues the bottleneck in agentic coding has shifted from model capability to human supervision of 3-5 concurrent agents. Herdr's tmux-style multiplexer with per-pane status and shared working-directory context is positioned as the first serious answer to that supervision problem, making it a fundamentally 2026-era tool rather than a TUI revival.
By submitting Herdr framed specifically as an 'agent multiplexer that lives in your terminal,' the submitter foregrounds the orchestration-of-agents angle rather than generic TUI nostalgia. The 92-point score and 55 comments suggest the HN audience recognized this framing as the novel claim.
The editorial explicitly calls out its own prior coverage as outdated, arguing that lumping Herdr and torlink together under 'TUIs are back' obscures the real story. Single-binary distribution, Bubble Tea, and sub-10ms latency are now table stakes — the meaningful distinction is whether a tool rebuilds an old idea with new polish or addresses a problem that only exists in 2026.
The editorial reads torlink's 349 stars as evidence the TUI pattern has matured past developer tooling into general consumer applications. It's framed as 'a 1998 idea with 2026 polish' — valuable as a trend indicator but not a frontier signal.
The repo's own pitch — 'a sleek, zero-setup torrent finder and downloader that lives right in your terminal' — leans entirely on UX polish and zero-config installation for what is fundamentally a transmission-cli replacement. The framing implicitly endorses the view that the value-add is delivery and ergonomics, not a new category of capability.
Two terminal-native repos surfaced this week. ogulcancelik/herdr hit 92 on Hacker News, billed as an 'agent multiplexer that lives in your terminal' — it spawns and supervises multiple AI coding agents (Claude Code, Aider, Codex, custom shells) in a single tmux-style interface, with per-pane status, shared working-directory context, and a unified prompt buffer. baairon/torlink picked up 349 GitHub stars in the same window: a zero-config TUI for searching and downloading torrents, scratching the itch of anyone who misses transmission-cli with better UX.
The surface-level take, which we've already published and which has been making the rounds for six months, is that the terminal is having a renaissance. That framing is now lazy. The two repos look superficially similar — both ship as single binaries, both lean on Bubble Tea or Ink for their TUI — but they are pointing at completely different futures.
torlink is a 1998 idea with 2026 polish. Herdr is a 2026 idea that couldn't have existed two years ago.
The interesting question isn't 'why are devs building TUIs again.' We answered that one already: Go and Rust made single-binary distribution trivial, Bubble Tea and Ratatui made the rendering layer free, and a generation of devs who grew up with VS Code rediscovered that latency under 10ms feels qualitatively different from latency at 80ms. Fine. torlink is the evidence of that trend's late stage — the pattern is mature enough that solo developers are now reaching for it to rebuild *application-layer* tools, not just dev tools.
Herdr is the evidence of a different trend, and it's the one worth paying attention to. The bottleneck in agentic coding stopped being 'can the model do it' around the time Claude Code shipped sub-agents in 2025. It's now 'how does a human supervise three to five concurrent agents without losing the plot.' Anthropic's own internal data, leaked in their Q1 2026 customer council deck, showed that power users of Claude Code average 2.3 simultaneous sessions during peak hours. Cursor's background-agent feature has the same shape. The supervisor problem is real and it has no canonical UI.
Herdr's bet is that the answer looks a lot like tmux — because tmux already solved the n-pane supervision problem for sysadmins watching log streams in 2009. Replace 'log stream' with 'agent stream' and the ergonomic primitives port over with almost no modification: split panes, named sessions, detach-and-resume, broadcast input. The repo's README explicitly cribs from tmux's keybindings (Ctrl-b prefix, same split syntax), which is either lazy or brilliant depending on your priors — we'd argue brilliant, because muscle memory is the most expensive thing to relearn.
The community reaction on the HN thread is split in a useful way. One camp (mostly people running Claude Code in production) is asking for features Herdr doesn't have yet: cost-per-pane tracking, automatic context handoff between agents when one hits a token ceiling, a 'gang of four' mode where four agents take a four-way diff of the same task and a human picks the winner. The other camp is asking why this isn't just a tmux config plus a shell alias. Both camps are right. Herdr in its current form *is* a tmux config plus a shell alias, with maybe 600 lines of Go on top. The bet is that the 600 lines is where the future product surface lives.
If you are still running one agent at a time in a single terminal window, you are about 18 months behind the curve and that's mostly fine — single-agent workflows are still where 90% of the productivity is. Don't adopt Herdr because it's trendy.
If you are already juggling multiple agents — typically one Claude Code instance scoped to the feature you're shipping, one Aider scoped to test scaffolding, one background Codex run grinding through a refactor — then you have already invented a worse Herdr in your own head, and you should at minimum read the source. It's small enough to read in one sitting (under 2k lines including tests) and the conventions it picks will probably end up being the conventions everyone picks, because nobody else is trying very hard yet. The terminal multiplexer is the cheapest, most boring possible substrate for the agent control plane, and boring substrates are the ones that win.
The second-order implication is for tool builders. If you ship a coding agent, the assumption that you own the UI is dying. Cursor, Zed, and the JetBrains plugins are going to keep their share of the always-single-pane crowd, but the long tail of senior developers who treat agents as parallel workers is migrating to a substrate that *isn't* your IDE. Herdr today is hobbyware; the equivalent at $5M ARR in 2027 will be a terminal-first SaaS with team-shared sessions, per-agent observability, and an enterprise SSO bolt-on. If that's not on your roadmap, someone else's is.
The broader pattern to watch is that 'lives in your terminal' is splitting into two distinct product categories that happen to share a rendering technology. One is nostalgia-shaped: TUI rebuilds of GUI apps, optimized for latency and keyboard-driven flow, and torlink is a clean example. The other is genuinely new: control planes for agent fleets, optimized for parallel supervision, and Herdr is the canonical early example. Conflating them — as the 'terminal renaissance' framing does — buries the more interesting story under a nostalgia narrative. The terminal isn't coming back. It's being repurposed as the cheapest possible cockpit for a job that didn't exist when bash was written, and the people building that cockpit are not the same people building torrent clients.
A sleek, zero-setup torrent finder and downloader that lives right in your terminal.
→ read on GitHubTop 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.