Devin Desktop (Formerly Windsurf): What Changed

Devin Desktop is the new name for Windsurf, the AI code editor from Cognition. The rename took effect on June 2, 2026 as an ordinary over-the-air update: same editor, same extensions and keybindings, same plan and pricing. Three things did change. The app now opens on the Agent Command Center, a Kanban view of all your local and cloud agent sessions. Its built-in agent is Devin Local, which shares its harness with Devin CLI and replaced Cascade; Cascade was removed on September 8, 2026. And the editor can host third-party agents, such as Codex CLI and Anthropic’s Claude Agent, through the Agent Client Protocol. If you used Windsurf, you already have Devin Desktop.

What the rename kept, and what it moved

Cognition’s announcement, Windsurf is now Devin Desktop↗ (opens in a new tab), calls it a full IDE with an agent manager built in, backward-compatible with Windsurf and VS Code. The editor, LSPs and extensions are where they were, and existing users got the update without changing plans. What moved is mostly names and folders:

• The application is now called Devin: Devin.app on macOS, Devin.exe on Windows. Device-management policies that allow only “Windsurf” may block it, so IT teams need to add the new name.
• The surf shell command became devin-desktop. The old surf and windsurf commands still ship for backward compatibility.
• On first launch, Migrate from Windsurf copies your settings, keybindings, snippets, workspace state and extensions to the new Devin data folders. Caches, logs and local history are skipped.
• In a repository, .devin/ is the preferred folder for rules, skills and plans, with .windsurf/ still read as a fallback. The single .windsurfrules file and AGENTS.md are read too.
• Legacy Windsurf Enterprise customers still manage settings and sign in at windsurf.com; documentation and the changelog moved to docs.devin.ai.

Devin AI vs Devin Desktop

Devin is now a brand covering four products, and the one people mean by “Devin AI” is usually the cloud agent:

• Devin Cloud: the autonomous agent that works on its own machine in the cloud, started from its web app, Slack, Linear or Jira. It is compared with Anthropic’s agent in Devin vs Claude Code.
• Devin Desktop: the IDE formerly called Windsurf, with Devin Local as its local agent and a Command Center that also shows Devin Cloud sessions.
• Devin CLI: a terminal agent built on the same harness as Devin Local.
• Devin Review: code review on pull requests.

You do not need Devin Cloud to use Devin Desktop. Cognition’s FAQ says you can keep working with local-only agents, and that Devin Cloud remains a separate product. The self-serve plans are Free, Pro, Max and Teams, with Enterprise above them. On Teams, only full seats include Devin Desktop; flex seats do not.

Devin Local replaced Cascade

Cascade was Windsurf’s agent, and most Windsurf guides are about it. Its successor, according to the Devin Local documentation↗ (opens in a new tab), is “our next-generation agent harness shared with Devin CLI.” Cognition’s announcement says it was rewritten from scratch in Rust, and the vendor’s own figure is up to 30% fewer tokens than Cascade for most tasks. What it brings:

• Modes: Normal, Plan and Ask. Plan mode researches read-only and writes its plan to ~/.devin/plans/plan-<session>.md before asking to implement.
• Subagents, behind a Subagents (Preview) toggle, including custom ones defined as Markdown files under agents/.
• An optional OS-level sandbox that limits writable paths and filters network domains.
• Worktree sessions: run the agent in a new or existing Git worktree and click Merge when it is done.
• Permissions instead of auto-execution levels: Deny, Ask and Allow rules, with Deny winning, scoped to file reads, file writes, commands, web fetches and MCP tools, and set per project, per user or by your organization.

It does not do everything Cascade did. Memories, workflows, app deploys and Arena mode are not supported. The command palette entry Devin: Open Cascade Migration Wizard moves memories, workflows and hooks into skills, which Devin Local loads when a task calls for them. On Enterprise plans Devin Local is off until an admin turns it on; on Teams and other plans it is on by default.

When Cascade went away, and where the docs disagree

Cognition’s pages do not tell one story. The June 2 announcement said Cascade stayed available “through July 1st,” and the FAQ says “through July.” The Devin Desktop changelog↗ (opens in a new tab) is later and more specific: version 3.9.19, dated September 8, 2026, says Cascade has been removed and Devin Local is now the only agent, with a Continue in Devin Local prompt for old conversations. Meanwhile the FAQ still says “Nothing is getting removed,” the getting-started page still links Cascade’s MCP, workflows and App Deploys pages, and the ACP page lists Cascade as a built-in agent. Treat the changelog as current and the rest as pages that have not caught up.

MCP configuration in Devin Desktop

Because Devin Local runs Devin CLI’s harness, it reads Devin CLI’s MCP files, not Cascade’s old ~/.codeium/mcp_config.json. The MCP configuration page↗ (opens in a new tab) gives three scopes:

• Project: .devin/mcp_config.json, checked in and shared with the team.
• Local: .devin/mcp_config.local.json, gitignored, for your own keys. This is the default when you add a server from the command line.
• User: ~/.config/devin/mcp_config.json, or %APPDATA%\devin\mcp_config.json on Windows, for every project.

Another disagreement to know about: the Devin Local page still shows the older layout, with servers under mcpServers in .devin/config.json and ~/.config/devin/config.json. The CLI page explains that the dedicated mcp_config.json files replaced those in version 3000.3, and that old entries are migrated automatically on startup, so either will work.

devin mcp add -s user fenbs https://fenbs.ai/api/mcp
devin mcp login fenbs

# ~/.config/devin/mcp_config.json then holds:
{
  "mcpServers": {
    "fenbs": { "url": "https://fenbs.ai/api/mcp" }
  }
}

Remote servers use Streamable HTTP and fall back to SSE when the server does not support it. Unlike Cascade, Devin Local asks before calling any MCP tool; you can allow one tool or every tool on a server, for the session or permanently. When a server’s sign-in expires it shows Needs auth, and Authenticate runs the browser flow again. Each client keeps its own OAuth session, so signing in from Claude Code does not sign in Devin.

Third-party agents over ACP

Devin Desktop is not only for Devin. Cognition’s ACP page↗ (opens in a new tab) lists Codex CLI, Anthropic’s Claude Agent, OpenCode, Junie and Gemini CLI as agents you can run inside the Command Center, where they get the same Kanban view and Spaces as Devin’s own. They are available on the Pro, Max and Teams plans, and Enterprise admins enable them through their account team. When you run one, Devin Desktop’s own terms do not apply and billing is between you and that agent’s maker. The Windsurf plugin for JetBrains IDEs is in maintenance mode, and Cognition recommends running Devin there over ACP instead.

Windsurf vs Devin Desktop: who it suits

• You liked Windsurf’s editor: nothing to decide. Devin Desktop is the same IDE, and your settings came across.
• You relied on Cascade memories or workflows: run the migration wizard and check the skills it writes before your next big task.
• You run several agents, local and cloud, and want one board of their sessions: the Command Center is the reason to be here, including for Codex or Claude’s agent over ACP.
• You want a terminal agent rather than an editor: Devin CLI runs the same harness, or compare Anthropic’s in Windsurf vs Claude Code.
• You manage laptops for a team: allowlist the Devin app name and the .devin folders, and move rules to .devin/rules/ when convenient.

One more recent change: since the September 29, 2026 release you can sign in to Devin with ChatGPT and have GPT model requests in Devin Desktop billed to an eligible ChatGPT Plus or Pro plan instead of your Devin usage.

A board for the work, not just the sessions

The Command Center tracks agent sessions: what is running, what is waiting, which pull requests came out. It does not hold the list of work that has not started yet, or the reasons behind it. fenbs does. It is a board shared by your team and your AI assistants, with lanes To Do, Next Up, In Progress and Completed, where each task carries a note with the problem, a plan and a test status. With the server added as above, Devin Local can read a task with fenbs_get_item, move it to In Progress, comment with what changed and move it to Completed, and History shows each change under the assistant’s name, on your behalf. The first call prompts for a browser sign-in; if an agent cannot complete it, issue a token under Settings, “Connect an AI assistant”, with a name, the scopes it needs and an optional expiry. fenbs has no sprints, due dates or settable assignee, so the lanes and a 1 to 10 priority carry the order.