loading
loading
Cognition Rebuilds Devin Desktop to Run Parallel Coding Agents on One Board Cognition ships a polished refresh of its agent orchestration IDE, free to download and now compatible with your existing ChatGPT Plus or Pro quota. - Cognition shipped a refreshed Devin Desktop with a polished Agent Command Center, free to download. - You can now bill OpenAI model usage to your existing ChatGPT Plus or Pro subscription. - Kanban-style dashboard tracks local and cloud agent sessions, PRs, and reviews in one view. - Agent Command Center now follows selected Spaces and supports side-by-side agent windows. - ACP support lets Codex, Claude Agent, and OpenCode run alongside Devin in the same board. - Built on the former Windsurf IDE, keeping extensions, keybindings, and LSPs intact. Cognition has refreshed Devin Desktop, its agent-oriented IDE, with a redesigned command center for coordinating local and cloud coding agents. The application is free to download, and developers can charge eligible OpenAI model usage to an existing ChatGPT Plus or Pro subscription instead of consuming Devin quota. Devin Desktop is the product previously called Windsurf. Cognition has retained the editor, extensions, keybindings, workflows, and language server support under the Devin brand, as its Windsurf FAQ explains. The latest release concentrates on the Agent Command Center, which manages parallel coding sessions through Spaces and a Kanban-style board. One board for every agent The Agent Command Center now serves as the default workspace for monitoring agents, reviewing their status, and moving between tasks. Developers can keep the full editor available while coordinating several sessions across local machines and Devin Cloud. Cognition’s release notes describe these changes as workflow improvements to the existing command center. The underlying Windsurf editing experience remains available inside the same application. ChatGPT billing, with boundaries Developers with ChatGPT Plus or Pro can authorize Devin to use that subscription for supported OpenAI model requests. The connection works across Devin Cloud, Devin Desktop, and Devin CLI, avoiding a separate Devin charge for eligible GPT usage. - Open Settings and select Connections. - Choose Continue with ChatGPT. - Authorize Devin to use the ChatGPT plan. - Enable Use your ChatGPT plan in settings or from a GPT model’s entry in the model picker. The selected model determines which quota is consumed. In Local Fusion on Devin Desktop and Devin CLI, choosing a GPT model for Fusion charges that model’s usage to the connected ChatGPT subscription. Requests to other model providers continue to consume Devin quota. Developers running mixed-model workflows should check the selected model before starting a session. The desktop application and its free tier lower the initial cost of testing the workflow, although model requests still count against the applicable ChatGPT or Devin allowance. Existing plan limits continue to govern how much GPT usage is available. ACP brings outside agents aboard Devin Desktop supports the Agent Client Protocol, or ACP, which provides a shared interface between coding agents and editors. Its role resembles the Language Server Protocol: an ACP-compatible agent can connect to an ACP-compatible editor without a custom integration for every pairing. Zed Industries created the open protocol, which JetBrains, Google, and GitHub have also adopted. Codex CLI, Claude Agent, OpenCode, and custom ACP-compatible agents can appear in the same command center as native Devin sessions. Spaces supply project-level context, and the board exposes a consistent set of controls for tracking work across those agents. A parallel workflow can divide one project into several coordinated tasks: - Assign a Claude agent to refactor a legacy module. - Send a Codex agent to write tests for the revised code. - Use Devin Local to handle integration work. - Track each session on one board under a shared Space. Where the workflow fits Devin Desktop is designed for developers who run several coding tasks concurrently and still need a complete IDE. The command center is most useful when agents work on separate branches, modules, or stages of a larger change and their status must remain visible in one place. - Multiple local and cloud agent sessions run at the same time. - Tasks need shared project context through Spaces. - Developers regularly switch between supervising agents and editing code. - Teams already pay for ChatGPT Plus or Pro and want to use that allowance for eligible GPT requests. Single-agent workflows gain less from the board because the editor already provides enough visibility for one active session. For parallel work, the refreshed command center addresses a concrete coordination problem: identifying which agent owns each task, where it is running, and how far it has progressed.
document fetched 2026-10-24 15:22