Run Devin CLI with your team in AQ

Published September 8, 2026 · by the AQ team

Devin CLI is Cognition's terminal coding agent, released on April 27, 2026: a local agent with full access to your codebase and tools, a custom terminal renderer written in Rust, a choice of frontier models plus Cognition's own SWE-1.6, and a hand-off that sends a session to a cloud agent with its own computer when work outgrows your laptop. AQ runs Devin CLI as a first-class harness: connect your Devin account once, and every workspace can start a Devin session on your team's always-on machine, in its own git worktree, with teammates able to open the same live terminal. This page covers what the CLI does, how the connection works, and what AQ adds, as of September 2026.

What Devin CLI is

Devin CLI (Cognition also calls it Devin for Terminal) is a single binary you run inside a project. You give it a task in plain language; it reads files, edits, runs shell commands, and checks its work. Three things distinguish it from the other terminal harnesses as of September 2026:

• Model choice at the prompt. Switch between frontier models and Cognition's SWE-1.6 with the model flag or the /model command, and /fast for SWE-1.6 Fast.
• Permission modes. Normal mode approves reads and asks before writes and shell commands. Accept-edits auto-approves file changes. An unrestricted mode auto-approves everything, and an autonomous mode runs inside an OS-level sandbox.
• Cloud hand-off. Start locally, then hand the session to a Devin cloud agent that keeps working on its own machine. This is Devin's answer to the closed-laptop problem.

# Install (as of September 2026)
curl -fsSL https://cli.devin.ai/install.sh | bash

# Sign in from a remote or SSH terminal
devin auth login --force-manual-token-flow

# Run in a project
cd your-project
devin

Connecting Devin CLI to AQ

Open Connections in AQ, choose Connect harness, and pick Devin CLI. AQ starts Devin's manual token flow in your per-user environment on the team machine and surfaces the approval link from app.devin.ai as a clickable banner, so sign-in works even though the CLI is running on a server with no browser. The token lands in your own home directory on that machine, not in AQ's database, and every teammate signs in with their own Devin account.

AQ installs the CLI for you. Every AQ machine gets a pinned Devin CLI build downloaded from Cognition's official release artifact and verified by SHA-256 during provisioning. That keeps every workspace on a team on the same version, and it means nobody runs an install script by hand on a shared machine.

What AQ adds

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. With Devin CLI connected, you get:

• An always-on session. Devin runs in a persistent tmux session on your team's VM (your own cloud, or a dedicated AQ-managed VM). Close the laptop, open the workspace from your phone, and the session is where you left it.
• One isolated worktree per workspace. Each workspace is a git worktree on its own branch, so several Devin sessions, or a Devin session next to a Codex session, never step on each other. Devin's own worktree-free multi-agent mode works inside that boundary.
• Multiplayer, not screen sharing. Teammates open the same workspace and see the same terminal. Typing into it is delegated: they request control and the owner approves. Comments pinned on the app preview can be sent to the agent as prompts.
• Trust prompts handled. Devin asks you to trust each new folder before it works there. Every AQ workspace is a new folder, so AQ pre-answers that for the worktree it created. Fresh workspaces never stall on the prompt.
• The rest of the workflow. Agents commit and open PRs under each user's own GitHub auth, PRs are tracked per workspace, and a Linear issue labeled ai-task becomes a workspace with ownership following the assignee.

Honest scope

Devin CLI runs in AQ as a terminal harness. That means the real CLI in a real terminal, with everything the CLI itself supports, including its permission modes, model picker, and cloud hand-off. It is not yet available inside AQ Composer, AQ's native conversation surface, because Composer relies on structured session events that Devin CLI does not publish today. Claude Code, Codex, Cursor, Kimi Code, Grok, Antigravity, Command Code, and OpenCode do run natively in Composer. If Cognition exposes that protocol, Devin joins them.

Billing is Devin's. You pay Cognition for Devin CLI on your own plan; AQ does not proxy or mark up model usage for any connected harness. Pricing for AQ itself is two plans: Free is a personal sandbox for one person (AQ creates a private machine, nothing to install, no time limit), and Team is $50 per user per month early access (standard $200, billed monthly), covering VMs you connect from your own cloud or a dedicated always-on AQ-managed VM.

When to pick Devin CLI in AQ

Pick Devin CLI when you want Cognition's models and its permission model, or when your team already pays for Devin and wants the terminal agent on shared infrastructure instead of on individual laptops. Run it side by side with Claude Code or Codex on the same task in parallel workspaces and keep the better diff. Our guide to choosing a coding harness covers the criteria, and the harness directory keeps the whole field in one place.