All field notes

Setup · 5 min read ·

Devin CLI: install, Windows, vs Claude Code

Install Devin CLI on macOS, Linux or Windows, sign in with devin auth login, see what Devin costs, and compare it with Claude Code, from Cognition’s own docs.

What is Devin CLI?

Devin CLI is Cognition’s coding agent for the terminal: it works on your local files, connects to Devin Cloud, and installs with one command on macOS, Linux, WSL or Windows. You sign in once with devin auth login, and Devin’s Pro plan, at $20 a month, includes a daily and weekly usage quota that covers the CLI.

Facts below come from Cognition’s Devin CLI docs and plans page, and the Claude Code side from Anthropic’s own pages, all checked on October 24, 2026. Both tools change fast, so check the docs before you rely on a flag. One gap to know up front: the quickstart says Devin CLI does not yet support Knowledge, Playbooks or Secrets from your Devin account.

How do you install Devin CLI on macOS or Linux?

Run the install script from the quickstart in a terminal. It works in WSL too:

curl -fsSL https://cli.devin.ai/install.sh | bash

On a Mac, Homebrew works as well: brew install --cask devin-cli, and brew upgrade --cask devin-cli later. Don’t run the script with sudo: Cognition’s troubleshooting page says that can cause ownership issues. Then open a new terminal, go to your project and run devin. If the command isn’t found, restart the terminal or run source ~/.zshrc (or ~/.bashrc); the same page gives ~/.local/bin/devin as an example location.

If you use Devin Desktop, Cognition’s editor and the new name for Windsurf, per its page, the CLI comes bundled: open the Command Palette and run Install Devin CLI. The quickstart says that route is for Legacy Windsurf Enterprise and Devin Enterprise plans, once an admin turns it on in team settings.

How do you install Devin CLI on Windows?

It runs natively, no WSL needed. Open PowerShell, not CMD or Git Bash, and run:

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

The quickstart also links standalone installers for x86_64 and ARM64 Windows. Once installed, devin works from PowerShell, Windows Terminal or Git Bash.

Three fixes from the troubleshooting page. An execution policy error: run Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy RemoteSigned, then the install again. “irm” or “iex” not recognized: you are in Git Bash or CMD, and those two are PowerShell aliases. devin not found afterwards: restart PowerShell, then check your PATH with $env:PATH -split ';'.

One Windows difference matters once it runs. On macOS and Linux, Devin takes a snapshot of your login shell’s environment at startup, so tools from nvm or pyenv are there; on Windows that snapshot doesn’t apply, so the tools your project needs must be on the system PATH.

How do you sign in, and what does Devin cost?

Run devin auth login and finish in your browser. On a remote machine or over SSH, devin auth login --force-manual-token-flow lets you paste a token instead, and devin auth status shows where you stand (command reference). Cognition says CLI API tokens don’t expire by default.

Plans, from Cognition’s plans page as of October 24, 2026:

Free: limited Devin usage, with Devin Review and DeepWiki.

Pro, $20 a month: a daily and weekly usage quota that covers Devin sessions, Devin CLI and Devin Desktop, and pay-as-you-go usage past it.

Max, $200 a month: a much larger weekly quota, with no daily cap.

Teams: from $80 a month, with full seats at $40 a month each and a quota equal to Pro’s.

The allowances refresh on a calendar basis, and Pro’s daily allowance is more than a seventh of its weekly one. Inside a session, /usage shows the session’s estimated usage and /upgrade changes your plan.

Which models does Devin CLI use?

Models from several vendors. The models page names Anthropic, OpenAI, Google and Cognition models, with short names such as opus, sonnet, codex, gemini and swe that always point at the newest model in that family, plus open models such as DeepSeek, Kimi and GLM. /model switches during a session, devin --model sonnet picks one at the start, and devin models list prints the models you can use. Enterprise teams can restrict the list in Team Settings.

Devin vs Claude Code: what is different?

Both are terminal agents that read your project, edit files, ask before risky actions and run without the interface in scripts. The differences are the models, the price and Devin’s link to Devin Cloud. Claude Code facts are from Anthropic’s setup page, pricing, memory docs and checkpointing docs.

Devin CLI and Claude Code, from Cognition’s and Anthropic’s own docs, checked October 24, 2026
TopicDevin CLIClaude Code
MakerCognitionAnthropic
InstallA script for macOS, Linux and WSL, Homebrew, and a PowerShell script or installer on WindowsA script for macOS, Linux and WSL, Homebrew, WinGet, apt, dnf or apk, and a PowerShell or CMD script on Windows
Sign-indevin auth login, with a Devin accountclaude, then the browser, with a Pro, Max, Team, Enterprise or Console account
Lowest paid planPro, $20 a monthPro, $17 a month billed yearly or $20 billed monthly
Free planLimited Devin usageThe free Claude plan doesn’t include Claude Code
ModelsAnthropic, OpenAI, Google, Cognition and open modelsClaude models
Rules filesAGENTS.md, AGENT.md and CLAUDE.md, plus commands, subagents and hooks from .claudeCLAUDE.md, or AGENTS.md when a project has no CLAUDE.md
Approval modesNormal, Accept Edits, Smart, Bypass, and Autonomous inside a sandboxManual, Accept edits, Plan, Auto and more
Undo/revert <step> undoes file changes from that step on and rewinds the conversation/rewind restores code, the conversation or both
Scriptsdevin -p "prompt"claude -p "prompt"
Cloud/handoff moves the task to Devin Cloud, and /pickup brings its pull request branch backCloud sessions keep going when you close the app

So which should you use?

If you already pay for Claude Pro or Max, Claude Code is part of your plan. Devin CLI makes sense if you want one agent across several vendors’ models, or if you already use Devin Cloud and want local sessions that hand work to it and back. Because it reads CLAUDE.md and, by default, imports commands, subagents and hooks from .claude, a repository already set up for Claude Code needs no second rules file for Devin (extensibility docs).

You don’t have to choose blind. Give both the same task from the same commit, each in its own worktree, and let your tests decide; our bake-off guide has the steps and a scorecard. For other agents worth a look, see our list of Claude Code alternatives.

Run Devin beside Claude Code in SwarmPane

From version 1.3.13, released October 23, 2026, SwarmPane opens Devin in a pane like any other agent: it finds devin on your computer, or shows its install command in Settings → Agents. Sign in once with devin auth login. Settings → Usage then shows Devin’s daily and weekly limits, read from Devin itself, and a Devin pane’s usage meter shows Devin’s own usage. SwarmPane never reads Devin’s sign-in (changelog).

That puts Devin next to Claude Code, Codex and the rest in one window, each pane with the status its CLI reports, and each in its own worktree if you want. Devin and Hermes Agent need SwarmPane 1.3.13, out now on Mac and Windows; Linux follows shortly. Cognition bills your Devin plan; SwarmPane is Starter at $17 a month, Pro at $29 or Ultra at $50. Start with a 7-day trial for $1 and cancel before it ends if it is not for you.

A Claude Code pane in SwarmPane with its account menu open, listing several saved accounts, and a weekly usage meter at the top of the window
Each pane has an account menu, and a usage meter sits beside the pane status: here Claude Code, at 27% of its week. From 1.3.13, a Devin pane’s meter shows Devin’s own usage.