Devin brings AI to terminal
- Cognition on April 27 released Devin for Terminal, a command-line version of its coding agent that runs on local machines and hands jobs to Devin Cloud. - Cognition says developers can start in a shell, then move the same session to a cloud agent that keeps working, opens pull requests, and resolves review comments. - The launch extends Devin beyond Slack, the web, and Visual Studio Code as agent tools shift toward mixed local-cloud workflows. (cognition.ai)
Coding agents are software tools that read files, run commands, and edit code for you. On April 27, Cognition brought its Devin agent into the command line with Devin for Terminal. (cognition.ai) (cli.devin.ai) Cognition says Devin for Terminal runs on a developer’s own machine, inside the shell they already use, with access to local code, tools, and environment. The company published an install command and says setup takes about two minutes. (cognition.ai) (cli.devin.ai) The product’s pitch is a split workflow: do fast, interactive work locally, then hand the same session to a cloud agent when the task gets bigger than a laptop. Cognition says the cloud agent runs on its own computer and can keep working after the user closes the terminal. (cognition.ai) That handoff is the part Cognition is emphasizing. In its launch post, the company says a cloud Devin can test code in its own browser, open a finished pull request, and even respond to review comments after the task leaves the local machine. (cognition.ai) In plain terms, the local terminal agent is the quick chat at your desk, while the cloud agent is the teammate who keeps working after you log off. Cognition is trying to combine both in one session instead of making developers switch tools mid-task. (cognition.ai) The release also adds another interface to a product Cognition had already pushed into Slack, web sessions, and a Visual Studio Code extension. When Devin became generally available in December 2024, Cognition priced it at $500 a month for engineering teams and highlighted Slack, IDE, and API access. (cognition.ai) Cognition has spent 2026 building out the cloud side of that system. On March 19, the company said Devin could manage multiple child Devins in parallel, each running in its own isolated virtual machine with its own terminal and browser. (cognition.ai) The terminal launch fits that strategy: keep the first step close to the developer, but push long-running work into managed remote machines. Cognition’s docs describe Devin for Terminal as a separate local tool with deep Devin Cloud integration, while noting that some cloud features, including Knowledge, Playbooks, and Secrets, are not yet available in the terminal product. (cli.devin.ai) Cognition also says developers can choose among models including Opus 4.7, GPT-5.5, and its own SWE-1.6 inside the terminal client. The company says it built the interface with a custom Rust terminal rendering library to keep it responsive. (cognition.ai) The bet is that developers want agents where they already work: in the shell, on local files, with a path to offload heavier jobs without rebuilding context. Devin for Terminal is Cognition’s latest attempt to make that handoff feel like one continuous session. (cognition.ai)