Devin CLI: how to install it, the main commands, and local vs cloud

What is the Devin CLI?

The Devin CLI is a local agent. Cognition's quickstart says it "runs directly in your terminal" and works with your local files and environment. It reads your code, edits files and runs shell commands, asking for approval based on the permission mode you pick.

It's one of the four products under the Devin name, next to Devin Cloud, Devin Desktop and Devin Review, and What is Devin AI? covers the family. The CLI isn't the same tool as Devin Cloud, which works in its own virtual machine. The quickstart says the CLI doesn't yet support the Knowledge, Playbooks and Secrets from your Devin account. Cognition's product page adds that it's built in Rust, and that MCP servers, skills, hooks and Claude Code plugins work in it as they are.

How do you install the Devin CLI?

You install it with one command for your system. These are the commands from Cognition's quickstart:

# macOS, Linux or WSL
curl -fsSL https://cli.devin.ai/install.sh | bash

# Homebrew (macOS)
brew install --cask devin-cli

On Windows, the quickstart links an installer for x86_64 and ARM64 PCs, or you can run this in PowerShell:

irm https://static.devin.ai/cli/setup.ps1 | iex

That command fails in Git Bash or CMD, so install from PowerShell. After that, devin works in PowerShell, Windows Terminal and Git Bash.

On the Legacy Windsurf Enterprise and Devin Enterprise plans, there's a fourth way: the CLI comes bundled with Devin Desktop. An admin first turns on Install Devin CLI in Devin Desktop in the CLI team settings. Then you open the Command Palette (Cmd+Shift+P on macOS, Ctrl+Shift+P on Windows and Linux) and run Install Devin CLI, which puts devin on your PATH.

How do you start the Devin CLI?

Restart your terminal, go to a project directory and type devin. That opens an interactive session where you chat with the agent and watch what it does. Cognition's essential commands list the ways to start it:

devin                          # Start an interactive session
devin -- your prompt here      # Start a session with a first prompt
devin -p "prompt"              # One turn, print the answer and exit
devin -c                       # Continue the last session in this directory
devin -r                       # Pick from recent sessions

The -- tells the CLI that the words after it are a prompt, not a subcommand. The -p mode is the one for scripts and CI. If you aren't signed in, devin auth login or /login connects the CLI to your Devin account.

What are the main Devin CLI commands?

The CLI has flags you pass when you start it, subcommands you run from your shell, and slash commands you type inside a session. These come from Cognition's commands and flags reference, which is also on your machine as man devin.

How much the agent does without asking depends on the permission mode. Normal, the default, approves read-only tools and asks before writes and shell commands. Accept Edits approves file edits in the workspace. Smart lets a fast model decide which other actions are safe, and it's still rolling out. Bypass approves everything, file edits included. Autonomous, which needs --sandbox, runs shell commands and fetches without asking because the sandbox limits what they can reach, but file edits through the edit and write tools still ask. Shift+Tab cycles through them.

Which models does the Devin CLI use?

The CLI runs models from Anthropic, OpenAI, Google and Cognition, plus open-weight models such as DeepSeek, Kimi and GLM, according to its models page. Short names like opus, swe and codex always point to the newest model in that family. Cognition recommends Fusion for most people, which pairs a frontier lead model with a cheaper sidekick.

How does the Devin CLI work with Devin Cloud?

The CLI can start, steer and resume Devin Cloud sessions. Its cloud docs say devin --cloud runs the session on its own virtual machine, with a shell, a browser and copies of your repositories, and streams it into your terminal. Closing the terminal doesn't stop it.

devin --cloud                                     # Start a cloud session
devin --cloud -p "fix flaky CI tests"             # One prompt, then exit
devin --cloud -r https://app.devin.ai/sessions/…  # Resume by URL

Handoff works in both directions. In a local session, /handoff creates a cloud session with your repo and branch, the conversation so far and your uncommitted changes, so commit or stash anything you don't want sent. Cognition suggests it for work that needs a server, a browser or a long run. In a cloud session, /handoff or /pickup checks out its pull request branch and starts a local session on it.

Which Devin plans include the CLI?

The CLI shares one quota with the rest of Devin. Cognition's self-serve plans docs say the Pro quota "covers Devin sessions, Devin CLI, and Devin Desktop", Max shares its larger weekly quota the same way, and a Teams full seat gets a Pro-sized quota. On Enterprise, the Devin auth docs say CLI usage is billed in ACUs against the organization's allocation, and admins give access with a role that has the Use Devin CLI permission. Devin pricing explains the quotas, extra usage and Teams seats.

Where Buildful fits

The Devin CLI runs in your terminal, on your machine, with Devin Cloud there when you want to hand off. Buildful is a background coding agent that only runs in the cloud. You describe a task in the browser, Slack or Linear, and it works in a private cloud sandbox, runs your checks, tests UI changes in a real browser and opens a pull request on GitHub for you to review.

Smart Mode picks the right open-weight model for each task, and on Pro and Team you can pick a frontier model like Claude or GPT yourself, with prepaid credit or your own API key. Buildful Pro is $10 a month for 1,000 tasks, or ₹750 in India, and Team is $40 a month for 3,000 tasks shared by unlimited members (₹3,500 in India). See Buildful vs Devin for the two side by side.

Frequently asked questions

Is the Devin CLI free?

Installing it costs nothing. Cognition's billing docs list the CLI in the usage quota of Pro at $20 a month, Max at $200 and Teams full seats. The Free plan has a light quota, and the docs don't say whether the CLI can draw on it.

How do I update the Devin CLI?

Run devin update, or /update inside a session, and add --force to reinstall the latest version. If you installed it with Homebrew, run brew upgrade --cask devin-cli instead.

Does the Devin CLI work on Windows?

Yes. Use the x86_64 or ARM64 installer, or run irm https://static.devin.ai/cli/setup.ps1 | iex in PowerShell. The --sandbox option doesn't run on native Windows, so Cognition says to use WSL 2 for it.

Is the Devin CLI the same as Devin?

No. The CLI is a local agent that works on your machine. Devin Cloud works in its own virtual machine and has Knowledge, Playbooks and Secrets, which the CLI doesn't support yet. The CLI can start Devin Cloud sessions and hand work to them.

How is the Devin CLI different from Claude Code?

Both are coding agents for your terminal. The Devin CLI runs models from several providers, while Claude Code runs Claude models. Cognition says the CLI reads Claude Code's CLAUDE.md, skills and MCP settings from your project. Devin vs Claude Code compares the two products in full.