Herdr Deep Dive
The "tmux for AI Agents": Run Multiple Coding Agents (Claude Code / OpenCode / Codex, etc.) From One Terminal With Persistent Sessions Over SSH and a Mobile-Friendly UI Used by Engineers at JetBrains / Docker / Vercel / Google / NVIDIA / AWS; Rust Binary, No Electron
Herdr is a terminal-based AI agent multiplexer — run multiple AI coding agents (Claude Code / OpenCode / Codex, etc.) in parallel from a single terminal, with sessions persisted server-side (like tmux) so you can reattach from any device. Key features: (1) multi-agent orchestration — run multiple agents in independent panes, (2) persistent sessions — agents keep running after terminal detach, reattach from anywhere, (3) SSH-friendly remote access — like tmux but agent-aware, (4) real terminal panes — clickable layouts and mouse support, (5) at-a-glance agent state (blocked / working / done / idle), (6) a Control API (CLI + JSON socket) for programmatic session management, (7) a mobile-friendly UI for terminal access from phones and tablets. Ships as a Rust binary (not Electron) with no external dependencies. Target audience: engineers at JetBrains / Docker / Vercel / Google / NVIDIA / AWS. Position: paired with Crit.md's HITL review substrate, Herdr is the "agent-operations infrastructure" side of the coin — solving the execution environment rather than the review loop. Part of the broader "day-to-day AI-agent tooling" trend alongside our Cloudflare-only stack and Claude Cowork web/mobile coverage.
TL;DR — What Herdr Is
Herdr is a terminal-based AI coding-agent multiplexer — run multiple AI agents from one UI, with sessions persisted like tmux.
Four takeaways:
1. tmux for agents — run Claude Code / OpenCode / Codex, etc., in independent panes
2. Persistent sessions — agents keep running after the terminal closes; reattach from anywhere
3. SSH + mobile-friendly — operate agents even from a phone or tablet
4. Rust binary — not Electron, no external dependencies; used at JetBrains / Docker / Vercel / Google / NVIDIA / AWS
The Problem — Parallel Agent Operations Are Chaos Today
Reality of running AI coding agents in 2026:
- Each agent in its own terminal — Claude Code here, Cursor there, Codex in yet another window
- Close the terminal, kill the agent — SSH drops during a long task and you restart from scratch
- Hard to check from the road — no visibility into agent progress on a build server
- No unified agent-state view — you have to hop between windows to see what's working, blocked, or done
Herdr's answer: take tmux's "multiplex many shells" structure and reapply it to "multiplex many agents", adding agent-specific state (working / blocked / done / idle).
The Seven Core Features
1. Multi-Agent Orchestration
Run multiple agents in independent panes:
- Pane A: Claude Code on a frontend feature
- Pane B: OpenCode on backend APIs
- Pane C: Codex generating tests
- Pane D: Cursor CLI on refactors
Each pane runs independently; focus moves via pane switching just like tmux.
2. Persistent Sessions
Agents keep running after you detach:
- SSH drops mid-generation → the agent keeps going server-side
- Reconnect from home to see what finished
- Check progress from the hotel on an iPhone
The tmux resume experience, applied to AI agents — essential now that long-running tasks are normal.
3. SSH-Friendly Remote Access
tmux-like but agent-aware:
- SSH into a remote box (prod terminal, GPU cluster, internal dev server) → herdr attach
- All agent sessions show up immediately, state fully synchronized
- Clean separation between where you develop (Tokyo) and where the compute runs (GPU cluster)
4. Real Terminal Panes
Real terminals, not Electron wrappers:
- Clickable pane layouts, mouse support
- Full ANSI escape support (color, decoration, cursor control)
- Each agent runs in its own shell environment — no conflicts with existing CLI tooling
Opposite direction from web-based IDEs (Cursor iOS, Claude Cowork): keep the terminal-native experience.
5. At-a-Glance Agent State
Sidebar shows every agent's state:
- blocked — waiting on user input (e.g., a Crit.md review comment)
- working — executing, still processing
- done — finished, awaiting review
- idle — waiting
"Which agent do I attend to next?" becomes an instant decision — a foundational productivity win for parallel operation.
6. Control API (CLI + JSON Socket)
Programmatic management:
- Start an agent from the CLI: herdr new claude-code
- Integrate with outside tools via JSON socket API (Slack notifications, monitoring dashboards, etc.)
- Workflows like auto-spawning agents from Cloudflare Durable Objects or Flue Framework become natural
7. Mobile-Friendly UI
Optimized for terminal access from phones and tablets:
- Responsive UI, SSH → herdr attach from iPhone / Android
- Same direction as Cursor iOS and Claude Cowork mobile — extending the developer's workspace from "desk" to "anywhere"
- Emergency response on the road, progress checks on the commute, one last debug before bed
Tech Stack — Lightweight Rust Binary
Design choices:
- Rust backend (src/app/state.rs, src/ui/sidebar.rs, etc.)
- Single-binary distribution, no external deps
- Not Electron — memory footprint measured in tens of MB (Electron-based tools regularly hit 500 MB–1 GB)
- Multi-agent integration — first-class support for Claude Code, OpenCode, Codex, Cursor CLI, and other major CLI agents
Installation: drop the Rust binary in place, run herdr. Customization via a config file similar to .tmux.conf.
Adoption and Target Users
Companies using it (from the official site):
- JetBrains (IntelliJ, PyCharm, etc.)
- Docker
- Vercel
- Google
- NVIDIA
- AWS
- and others
Forward-leaning dev teams where parallel agent operation is already routine — plus individual developers up through large teams.
Positioning — Paired With Crit.md
Day-to-day AI-agent operations infrastructure:
| Tool | Role |
|---|---|
| Crit.md | HITL review substrate — quality check on agent output |
| Herdr (this column) | Execution substrate — parallel operations and persistence |
| Claude Cowork | The agent itself — background execution + HITL notifications |
| Cursor iOS | Agent UI — mobile-first |
| Nous Portal | Model substrate — unify 300+ frontier models |
Herdr's uniqueness: the only tool that tackles "one person operating many agents at once" — the same problem tmux solved for "many shells" twenty years ago, redefined for the AI-agent era.
How People Use It
(1) Solo dev, parallel projects: front / back / infra as three panes, a different agent for each.
(2) Long-running tasks: hand a multi-hour task ("fix every type error in the monorepo") to Herdr and go out; check progress from the phone.
(3) Team dev servers: each engineer keeps their own Herdr session on a shared dev box, attaches over SSH.
(4) GPU-cluster operations: manage agents running Qwen 3.6-35B or Gemma 4 on high-end GPU nodes.
(5) Ops monitoring: keep Cloudflare / AWS monitoring agents running under Herdr, respond immediately from mobile when something fires.
Update (Oct 2026): The herdr-app Desktop App
In October 2026 Herdr's desktop app was published on GitHub as herdrdev/herdr-app, and on October 30 it reached the Hacker News front page with 365 points. Like the terminal version, it is built for running several coding agents side by side: it groups work into spaces, shows in a sidebar which agents need attention, and keeps terminals running after you close the window. A separate component, herdr-core, runs the terminals, and the installers add it if it is missing.
| Item | Details (from the README, as of Nov 2, 2026) |
|---|---|
| Platforms | macOS (Apple Silicon; Intel requires building from source), Linux (x86-64 and ARM64; needs a desktop session and a Vulkan driver), Windows (x64, Windows 10 1803 or later) |
| License | Apache-2.0 |
| Main features | Spaces, tabs and split panes; agent status (Working / Blocked / Done / Idle); work on SSH machines; custom layouts; JavaScript "mods"; a command palette (Ctrl+Shift+P, or Cmd+Shift+P on macOS); key bindings in keys.json; background updates |
| Notes | Usage telemetry is on by default in releases (turn it off under Settings → Privacy). There are no tagged releases on GitHub Releases yet |
# macOS / Linux
curl -fsSL https://releases.herdr.dev/app/install.sh | sh
# Windows (PowerShell)
irm https://releases.herdr.dev/app/install.ps1 | iexOn macOS (Apple Silicon) you can also drag and drop from Herdr.dmg. If you roll it out to company machines, decide on the telemetry setting first. For reviewing diffs, see Hunk, which is often paired with Herdr, and OmniDiff, which compares changes at the syntax level.
Caveats and Warnings
(1) Terminal-culture ramp: tmux-like UI means a learning curve for people without a tmux / screen background. GUI-first developers may prefer Cursor / Cowork.
(2) Security considerations: persistent sessions live server-side and hold agent state, so secrets like API keys need care. Restrict access to the Herdr server.
(3) Agent dependencies: expect to keep pace with agent API changes. Community activity will decide long-term health.
(4) Resource consumption: many agents in parallel add up in CPU / memory / LLM-API cost. Budget alerts belong in the design.
(5) Team governance: when multiple engineers share one Herdr host, session isolation and permission control need thought.
Bottom Line
Herdr is essential infrastructure for the era of parallel AI-agent operations. The tmux idea — solving multiplexing for shells — recast for AI agents. Persistent sessions + SSH + mobile UI let engineers operate many agents, from anywhere, continuously. Combine it with Crit.md (HITL review), Nous Portal (model substrate), and "just Cloudflare it" (infra) and you have a default stack for late 2026.
Related services from us — software development, AI consulting, Hermes Agent setup, and OpenClaw setup. For enterprise AI-agent parallel-operations environments, SSH-server design, or team governance, get in touch. For Herdr Annotate, which lets you annotate agent output inside herdr and send it back, see our Plannotator guide.
References
- Herdr
- Related columns: Crit.md (HITL review), Claude Cowork Web/Mobile expansion, Nous Portal Cloud, Cursor iOS support, Claude Fable 5 return
株式会社オブライト
Oflight Inc. is a Tokyo-based IT company providing software development, digital signage, web development, network systems, and IoT/AR/VR technology services.
Related free tools (no sign-up, instant results)
Feel free to contact us
Contact Us