AQ

AQ is a multiplayer development environment for teams that work with AI coding agents. Everyone opens the same workspaces in a browser, sees who's there, watches the agent's terminal live, and can take the keyboard when a task needs a human. It runs on your own infrastructure or on machines AQ manages for you, and it supports the agent CLIs your team already uses rather than replacing them with its own. The site's own description is a multiplayer agentic development environment that runs on your infrastructure, and each of those words is doing work.

The problem it goes after is that coding agents were built for one person at one terminal. When a team adopts them, the work scatters into private sessions nobody else can see, agents trample each other's files in the same checkout, and a laptop lid closing kills a half-finished job. AQ moves the sessions to persistent machines, gives every task its own isolated space, and puts the whole thing on a shared screen. It also answers the manager's question of what the agents are actually doing across a team, which today is usually reconstructed from pull requests after the fact.

The mechanics are concrete. Each workspace gets its own git worktree with a separate working directory and a branch named after the task, so it shares the repository's history but never collides with anyone else's files. AQ installs dependencies, rebases onto main on demand, and tears the worktree down when the workspace closes. Agent sessions run as persistent terminals on the runner and stream to the browser, so they keep going after you shut your laptop and pick up where they were when you return. Live sessions show presence and allow keyboard control to be handed between people, the code editor sits alongside the terminal in the same tab, and app previews open next to the code with comments and annotations so a designer or product manager can point at what's wrong instead of describing it. Workspaces can be created by hand, from a GitHub pull request or branch, or from a Linear issue tagged with a label, with status syncing back in both directions and the agent committing, pushing, and opening pull requests under each user's own GitHub access.

It's for engineering teams, and specifically for the mix of engineers, product managers, and designers who now all touch agent-driven work. Startups, remote teams, and agencies are the named audiences, along with engineering managers who want to see what the agents are doing across a team rather than guessing from the pull requests. It works from any device with a browser, including macOS, Windows, Linux, and mobile, which matters when the person approving a change is on a phone.

Two choices make it different. First, it's agent agnostic. Claude Code, Codex, Cursor Agent, Kimi, Grok, Antigravity, and plain shell sessions all run inside it, and you sign each one in with your own account, so there's no markup on model usage and no lock-in to a house model. Second, it runs where you want. The free tier gives you a personal sandbox on an AQ-managed machine, and teams can point it at VMs in their own cloud or take a dedicated always-on machine from AQ. Both matter for companies that want agents working on private code without shipping that code to a new vendor. The Team plan bills a seat only after a person accepts an invitation and joins, and the free sandbox is a cheap way to see whether the model fits before committing a team.

The limits are those of an early product. Team pricing is explicitly early access, the docs don't yet list which cloud providers the bring-your-own runners are tested against, and the free sandbox sleeps when idle and takes around thirty seconds to wake. It's a browser product, so people attached to a local terminal and editor will be trading that for a shared screen, and the site doesn't mention an offline or desktop mode. The Linear integration is the only issue tracker named, and sign-in requires a work email. There's a sales contact for larger deployments, but no published support address on the site.

Access is freemium. The free plan is a solo personal sandbox with nothing to install, $5 in model credits, and the full set of agent CLIs, editor, previews, and integrations. The Team plan is $50 per user per month at the early access rate, locked for twelve months against a standard price of $200, with unlimited shared workspaces, unlimited seats billed only after a person joins, and the choice of a dedicated VM or your own cloud. You sign in with a work email and with your existing agent accounts, so the model bills go where they already go, and AQ's own charge is the per-seat fee for the shared infrastructure.