AQ vs Herdr: two ways to keep a fleet of coding agents running

Published September 7, 2026 · by the AQ team

Choose AQ if you want your team running coding agents together: shared live terminals teammates watch and steer from a browser on any device, one isolated git worktree per workspace, on a VM you control on the Team plan or a free personal sandbox to start. Choose Herdr if you are one engineer who lives in the terminal and wants a free, open-source runtime that keeps agents alive on a machine you already manage. Both keep agents running after your laptop lid closes; the difference is that Herdr is a single-player terminal multiplexer you assemble around your own machines and SSH access, while AQ is a browser-based platform where the machines, the worktrees, and the whole team come built in.

Herdr moves fast: it announced it was joining Y Combinator in August 2026 and ships frequent releases. Every Herdr claim below is as of September 2026, verified against its GitHub repository, release notes, and documentation.

AQ vs Herdr at a glance

What is Herdr?

Herdr (herdr.dev) calls itself the runtime your coding agents live on. Concretely it is a single Rust binary, installed with one command or Homebrew, that runs a background terminal server on your machine. Every agent gets a real terminal pane backed by a real PTY, so full-screen TUIs, interactive prompts, and mouse support behave as they would in a bare terminal. The server keeps panes alive when your terminal closes, your SSH connection drops, or every client quits; you reattach later from the same machine, another machine over SSH, or a phone with any SSH client. As of September 2026 it runs on macOS, Linux, and Windows (Windows reached general availability in v0.8.2).

The feature that made Herdr take off is agent state awareness. It auto-detects 15+ coding agent CLIs (Claude Code, Codex, Copilot CLI, Cursor Agent, Gemini CLI, Devin CLI, Grok CLI, Amp, OpenCode, and more) and shows each pane's state in a sidebar: working, blocked on input, idle. Worktree commands wrap git worktree so each branch checkout gets its own workspace, and a CLI plus socket API lets scripts, or agents themselves, spawn panes, submit prompts, and wait on other agents. A community plugin marketplace passed 500 plugins in its first month, though listings are automatic and unreviewed.

The project is open source under Apache-2.0 and moving fast: it reported 25k GitHub stars and 340k downloads four months after launch, and announced in August 2026 that it was joining Y Combinator. The founder has said the runtime stays free and open, with the business to be built on features connecting multiple clients, VPS hosts, and sandboxes; as of September 2026 none of those are shipped or priced.

What is AQ?

AQ is the multiplayer coding harness where engineering teams run AI coding agents like Claude Code and Codex together: shared live terminals, a code editor, and app previews, in your own cloud.

Every AQ workspace is an isolated git worktree on a machine dedicated to you: on the Team plan a VM you connect from your own cloud or a dedicated always-on AQ-managed VM, on the Free plan a private sandbox machine AQ creates for you, nothing to install and no time limit. The agents inside it are the real CLIs your engineers already subscribe to (Claude Code, Codex, Cursor Agent, Kimi, Grok, or plain shells), running in persistent tmux sessions streamed live to the browser. Sessions survive a closed laptop and resume from any device, including a phone. Teammates open the same workspace and watch the same live session; the owner can delegate terminal input by approving a control request in one click. Live dev-server previews get shareable links that work without an account for viewing, and comments pinned on a preview can be sent to the agent as prompts. Work arrives by hand, from a prompt, or from Linear: label an issue ai-task and a workspace appears in the team sidebar, with status syncing both ways. Each engineer signs into the CLIs with their own accounts, and AQ never marks up model usage.

Where Herdr stands out

• Free and open source. The runtime is Apache-2.0, a small single binary with no account and no bill. If you already have a machine to run agents on, trying it costs one install command.
• A real terminal, natively. Real PTYs per pane, tmux-style keys, and zero webview between you and the agent.
• Agent state at a glance. The working / blocked / idle sidebar across 15+ auto-detected agents works with zero configuration.
• Agent-native automation. The CLI and socket API let agents drive the multiplexer itself: spawn a pane, hand off to another agent, wait for it to finish. Community plugins extend it further.
• Momentum. 25k stars and 340k downloads in four months, a fast release cadence, and YC backing suggest it will keep improving.

Where AQ wins

• The team is in the session, not just you. On AQ, teammates open the same workspace in a browser and watch the same live terminal while the agent works; the owner delegates input in one click. Visibility is owner-managed: team-visible, or private and shared with specific people. See multiplayer coding agents.
• The machine comes with it. Herdr keeps agents alive on machines you provision, patch, and pay for yourself; connecting VPS hosts and sandboxes is the part the company plans to charge for later. AQ's Team plan covers VMs you connect from your own cloud or a dedicated always-on AQ-managed VM in its own isolated network; the Free plan is a private sandbox machine AQ creates for you. There is no shared multi-tenant execution tier.
• A browser beats an SSH client for everyone else. Herdr on a phone means an SSH client and keys to a box. AQ is a URL: the same live terminal, editor, and app preview on macOS, Windows, Linux, and mobile, with per-user GitHub auth and per-user CLI logins handled in-product.
• Previews and outside reviewers. AQ runs a live dev-server preview per workspace with shareable links that work without an account for viewing; teammates pin comments (screenshots supported) that can be sent to the agent as prompts, and outside reviewers comment via a review link after a quick email sign-in. Design and product see the running app, not a scrollback buffer.
• Work intake from where work lives. Label a Linear issue ai-task and a workspace appears in the team sidebar; status syncs both ways, ownership follows the assignee, and humans start agents (never auto-launched). Agents commit, push, and open PRs, tracked per workspace. Herdr has no issue-tracker intake on its public surface as of September 2026.
• Worktrees without the ceremony. Herdr wraps git worktree with commands you run; AQ creates one isolated worktree per workspace automatically (branch ai/{id}-{slug}, auto dependency install, one-click rebase onto main), and empty repositories work. See running multiple coding agents in parallel.

The deeper difference: a runtime you assemble vs a place the team shares

Herdr and AQ agree on the premises this category now takes as given: agents should outlive your terminal, and one engineer will run many of them. Herdr answers as infrastructure for an individual: an excellent open-source primitive that keeps panes alive on machines you bring, with SSH as the door. Everything social about the work stays outside it. That matters because team adoption of agents is already lopsided: a July 2026 LeadDev analysis of 25,264 agent-generated pull requests across 2,361 popular GitHub repositories found that in 79 percent of agentic PRs the same developer both reviewed and modified the agent's contribution, and only about one in eight workflows involved multiple humans. A single-player runtime leaves that pattern intact: your agents are alive all night, and you are still the only person who ever sees them work.

AQ answers as a place: the machine, the worktree, the live session, and the team in one browser tab. A teammate does not SSH into your box to help; they click the workspace in the sidebar and watch the same terminal you do. If your bottleneck is keeping your own agents alive, Herdr solves it for free. If your bottleneck is that agent work is invisible to everyone but the person who ran it, that is the problem AQ was built around. For the persistence half alone, see keeping Claude Code running after closing your laptop and monitoring multiple agent sessions.

Pricing compared

Herdr's runtime is free and Apache-2.0; as of September 2026 its announced commercial features are unpriced, and you supply and pay for the machines agents run on. AQ's pricing is public: a Free plan (a personal sandbox for one person: AQ creates a private machine in an isolated network, nothing to install, no time limit) and a Team plan at $50 per user per month early access (standard $200, billed monthly, rate locked for your first 12 months) covering VMs you connect from your own cloud or a dedicated always-on AQ-managed VM in its own isolated network. Inviting a teammate costs nothing until the person accepts and joins. On both products, agents bill through your own Claude and OpenAI accounts; AQ never marks up model usage.

Bottom line

Herdr is the best open-source answer yet to "keep my agents alive and show me who needs attention", and terminal-first engineers should know it exists. But it is single-player by design: you bring the machines and the SSH access, and nobody else ever sees the session. AQ ships the other layer: dedicated machines on both plans, worktrees created per workspace automatically, and the whole team watching and steering the same live session from a browser on any device, with previews and Linear intake attached. One engineer herding their own agents will be happy with Herdr. A team that wants agent work visible, reviewable, and shareable while it happens is AQ's side of the line.