AQ

AQ is a browser-based multiplayer agentic development environment built by BetterLeap, Inc., currently in early access. It lets engineering teams — including PMs and designers — run AI coding agents like Claude Code, Codex, Cursor Agent, Kimi, and Grok together in shared live sessions, backed by dedicated infrastructure in your own cloud or an AQ-managed VM.

What It Is

AQ is a team-oriented coding harness, not an IDE plugin or a proprietary model wrapper. The core idea is that AI coding agent sessions should be shared, live places a whole team can enter together — watching, steering, and reviewing the same run — rather than single-player tools that produce diffs in isolation. Each workspace gives a task its own isolated git worktree, a persistent terminal session, a code editor, and a live app preview, all accessible from any browser on any OS.

How the Multiplayer Model Works

The workspace is the central unit. When a teammate opens a workspace, they see the same live terminal, the same editor, and the same preview as the person who started it — not a replay. Control can be handed off with one click: the recipient types as the session owner, the owner retains their own input, and control can be revoked at any time. This means when an agent needs a decision, the person with the answer can already be in the room rather than receiving a Slack summary after the fact.

• Parallel workspaces: Each task gets its own git worktree (ai/{id}-{slug}), so parallel agents never collide on files or dependencies.
• Persistent sessions: Terminals run in tmux on the VM, so closing a laptop doesn't stop the agent. Reopening the workspace from any device — including mobile — resumes the exact session.
• Guest links: People outside the team (founders, clients) can view a workspace preview without a seat.

Linear and GitHub Integration

Work enters AQ from Linear and exits as GitHub pull requests. Labeling a Linear issue ai-task automatically creates a workspace in the team sidebar and moves the issue to In Progress; status syncs both ways, and workspace ownership follows the Linear assignee. On the exit side, agents commit, push, and open pull requests using each user's own GitHub credentials, and AQ tracks every PR — opened, merged, and closed — on its workspace card.

Infrastructure and Deployment Model

AQ is explicit that execution happens on dedicated machines, not shared multi-tenant infrastructure. Teams can connect VMs from their own cloud account (AWS, GCP, etc.) for full custody of code, worktrees, and CLI credentials, or use a dedicated AQ-managed VM created with one click. The free sandbox is a private AQ-managed VM in an isolated network. The aq-runner doctor CLI command diagnoses runner issues, and runner health is visible in the product UI.

Supported Agents and API Surface

AQ supports Claude Code, Codex, Cursor Agent, Kimi, Grok, and Antigravity CLI, plus plain shell sessions. Each user connects CLIs with their own accounts, so model spend stays on the team's own subscriptions and AQ does not mark up usage. AQ also ships a built-in agent called AQ Composer that runs on AQ-hosted models with credits included per seat. Beyond the UI, the workspace lifecycle is scriptable via a versioned REST API, signed webhooks, and an MCP server at aq.dev/mcp that gives Claude, Cursor, or any MCP client the same surface with a single browser sign-in.

Current Status: Early Access

AQ is in early access as of mid-2026, with docs first published July 10, 2026 and updated September 12, 2026. Sign-up requires a work Google account and personal review by the AQ team, typically within a day or two. The product is described as moving fast, with early teams shaping the roadmap. The early-access pricing is explicitly promotional and locked for the first 12 months for teams that join during the promotion.