Windsurf

What you are actually buying, and what it is called now

If you are searching for Windsurf, you will land somewhere else. https://windsurf.com/ returns an HTTP 308 permanent redirect to https://devin.ai/desktop. The product was renamed Devin Desktop on 2 June 2026 and delivered as an over-the-air update to existing installs.

This is a rename and a repositioning, not a shutdown, and the vendor is unusually explicit about that. Cognition's transition FAQ states: "Devin Desktop is the new name for Windsurf. It's the same IDE, same editor, and has the same features, but unified under the Devin brand." It also says "Nothing is getting removed" and, on billing, "Your current plan continues to work exactly as it does today. Pricing is unchanged. This applies to all plan types, including legacy Windsurf Enterprise plans."

The Windsurf name survives in three concrete places, which matters if you are searching for support or downloads: the JetBrains plugin is still shipped and documented as "Windsurf for JetBrains"; legacy Windsurf Enterprise plans still exist as a billing category, and legacy customers are told they "can still use windsurf.com to manage your settings"; and the download CDN is still windsurf-stable.codeiumdata.com, a leftover from the original Codeium name. Only the windsurf.com root redirects — subpaths such as windsurf.com/support still return 200 directly.

The repositioning is real, though. The Agent Command Center (a Kanban view, Spaces, multi-agent management) is now the default surface rather than the editor, and Cognition says it expects developers to spend "90%+ of their days" there. If what you wanted was a quiet VS Code fork with good autocomplete, the product's centre of gravity has moved away from you.

The agent layer, and why it is the strongest thing here

The most genuinely differentiated feature is that Devin Desktop will run other vendors' coding agents inside itself, through the Agent Client Protocol (ACP) — an open protocol, documented at agentclientprotocol.com, that does for agents roughly what LSP did for language servers.

Cognition's own docs name the competition as examples: OpenAI's Codex CLI, Anthropic's Claude Agent, Google's Gemini CLI, JetBrains' Junie, and the open-source OpenCode. Any ACP-speaking agent plugs in, and you can write a custom one. A vendor documenting how to run its rivals' agents in its own editor is not the normal shape of this market, and if you are hedging against model or vendor lock-in it is a substantive reason to prefer this over a more closed editor.

Two caveats stated in the same doc, both fair rather than buried: ACP agents are available on Pro, Max and Teams but not Free, and Enterprise admins have to enable them. And when you run an external agent, "Devin Desktop's privacy policy and legal terms do not apply, and billing is directly between you and the third-party agent provider" — so an ACP agent is your relationship with that vendor, not Cognition's.

Switching costs are also deliberately low. Devin Desktop reads AGENTS.md, still reads legacy .windsurfrules and .windsurf/rules/, adds .devin/rules/, and can import Cursor's .cursor/rules (.mdc) files. Settings import from VS Code and Cursor is documented. That is a well-handled migration story in a category where most vendors would rather you rebuilt your config from scratch.

How pricing works, and the one number nobody publishes

Verified on devin.ai/pricing on 14 August 2026. Free: $0. Pro: $20/month. Max: $200/month. Teams: $80/month for the team plan plus $40/month per full developer seat, with unlimited additional "flex" seats. Enterprise: contact sales.

The seat model on Teams is worth reading carefully, because "unlimited team members" is doing some work. Per the pricing page tooltip, "Each full user includes their own generous quota and access to Devin Desktop" — so flex seats are free but do not get the desktop IDE. Access to the editor is what the $40 buys. The transition FAQ adds a genuinely favourable detail the pricing page omits: a full seat unlocks Devin Desktop "and $40/month of usage that works across all Devin products." So the seat fee is largely a usage deposit rather than pure rent.

In March 2026 the product replaced prompt credits with a quota system. Your plan includes a daily and weekly token budget; cost per token varies by model; some models are free and do not draw down quota at all. Once quota is exhausted, Pro/Max/Teams users can buy extra usage, and here the vendor is straightforwardly good: "Usage is billed at API list prices for the model you're using." Billing overage at list rather than a marked-up credit is more buyer-friendly than much of this market, and Cognition publishes a machine-readable per-model, per-tier price table (input, output, cache-read and cache-write cost per million tokens) inside the models doc, so overage really is calculable in advance.

What is not calculable in advance is the included allowance. Nowhere in the pricing page or the quota documentation is the daily or weekly quota stated as a number. The docs describe it only in relative terms — "Your daily quota is more than 1/7 of your weekly quota" — and the pricing page says only that for "the majority of users, this quota will be enough to fully cover all agent usage." You cannot work out how much agent work $20 buys until you are inside the product and can read the usage meter. That is a real gap for anyone budgeting a team rollout, and the honest advice is to trial it before committing.

Two further changes matter if you are comparing against older reviews. Existing subscribers are grandfathered indefinitely — Pro at $15/month, Teams at $30/seat — which means today's $20 and $40 are increases for new buyers only. And SSO has moved: "New Teams plans do not include SSO — it is now an Enterprise-only feature." If SSO is a requirement, Teams no longer clears the bar.

The extension trade-off: Open VSX, and what it means for C#, C++ and Unity

Devin Desktop is a VS Code fork, and like other forks it cannot ship Microsoft's proprietary extensions. Cognition states the consequence plainly in its setup guide: "Devin Desktop workspaces rely exclusively on open-source tooling for compiling, linting, and debugging. Microsoft's proprietary Visual Studio components cannot be redistributed, so we integrate community-maintained language servers, debuggers, and compilers instead."

In practice that means extensions come from Open VSX and Cognition's own marketplace.windsurf.com rather than Microsoft's marketplace, and specific stacks get substitutes:

- C# / .NET Core: a community C# extension (muhammad-sammy.csharp) bundling OmniSharp and NetCoreDbg, rather than Microsoft's official C# extension with Roslyn and vsdbg. - C / C++: clang-based tooling — clangd and LLDB — rather than Microsoft's cpptools. - .NET Framework: requires installing the Mono toolchain yourself, and the docs warn that the relevant OmniSharp build "does not support non-SDK-style .NET Framework projects, including Unity."

For most Python, TypeScript, Go and Rust work this is a non-issue; Open VSX covers those well. For a Unity shop or a legacy .NET Framework codebase, it is close to disqualifying, and it is the sharpest single reason a team might rule this tool out. It is not a hidden catch — Cognition documents it clearly and at length — but it is also not something you would discover from the marketing page, and it is exactly the kind of thing worth checking against your own stack before you buy seats.

Cascade or Devin Local: which agent you should actually use

There are now two first-party agents in the editor, and the naming has caused enough confusion that it is worth stating clearly: as of the v3.7.16 release notes and the current ACP documentation, both Cascade and Devin Local exist and both appear in the agent selector. Reports elsewhere that Cascade was retired on 1 July 2026 do not match Cognition's own documentation, which still maintains a full Cascade section covering modes, memories, workflows, hooks, MCP and Arena mode.

What has actually happened is a managed succession. Devin Local is the newer agent and the one Cognition steers you toward: the FAQ describes it as "more efficient than Cascade, supports subagents, and runs the same architecture as Devin CLI," and it inherits your existing Windsurf settings. Cascade is now positioned as legacy. On enterprise tiers it "defaults to disabled" as of the 7 August 2026 update, and since 31 July 2026 admins can scope it three ways — enabled everywhere, JetBrains-plugin-only, or disabled — described as a way to "move users to Devin Local while keeping Cascade in the JetBrains plugin."

The practical read: new work should target Devin Local, because it shares a harness with Devin CLI and Devin Cloud and is where the feature work is going. If you depend on Cascade-specific behaviour, check the Devin Local limitations page first, and if you are on an enterprise plan check whether your admin has already turned Cascade off.

Who owns this now, and whether it will still be here

The ownership history is unusually turbulent even by this market's standards, and it is the main thing a buyer should weigh on continuity.

The company began as Codeium. In April 2025 OpenAI was reported in talks to buy Windsurf for about $3 billion. That deal fell apart, and on 11 July 2025 Google paid $2.4 billion for a nonexclusive licence to certain Windsurf technology and hired CEO Varun Mohan, co-founder Douglas Chen and some senior researchers — taking no equity and no control. Most of the roughly 250-person team stayed. Three days later, on 14 July 2025, Cognition acquired the remaining company: IP, product, trademark, brand and staff, for undisclosed terms. So the people who built it largely left for Google, and the product is now owned by the company behind Devin.

On the evidence available, the acquirer is well funded and the product is not in maintenance mode. Cognition raised more than $1 billion in May 2026 at a $25 billion pre-money / $26 billion post-money valuation, led by Lux Capital, General Catalyst and 8VC, and was reported on 12 August 2026 to be in early talks at around $40 billion. Cognition says it reached a $492 million annualised revenue run-rate; that figure is the company's own and is not independently verified.

Shipping cadence is the more useful signal, and it is healthy. The desktop IDE released v3.7.16 on 10 August 2026, four days before this listing, following v3.6.27 on 1 August — roughly weekly. The Devin platform release notes are near-daily, most recently 12 August 2026. The Windsurf JetBrains plugin is slower but alive: v2.12.26 on 29 July 2026, though with month-long gaps earlier in the year, so treat the JetBrains plugin as maintained rather than actively developed.

The fair summary is that continuity risk is not about the vendor disappearing — it is that you are buying into a product whose original team has gone and whose roadmap now serves Cognition's agent platform strategy rather than the standalone editor it was.

What we could not establish

We found no credible independent review or benchmark of Devin Desktop. Searches surface a large number of pages that look like AI-generated affiliate content rather than hands-on testing, and we have not cited any of them. Cognition does not publish SWE-bench or comparable numbers for the in-editor agents on the pages we checked, and we did not run the product ourselves. Anyone comparing this against Cursor or Copilot on output quality is currently reliant on their own trial.

One live inconsistency in the vendor's own material, which we could not resolve and which affects cost: the quota documentation says SWE-1.7 is "free through August 8, 2026" and the models page says it is "available as a free preview until August 8" — both dates now passed. Yet as of 14 August 2026 the pricing page still advertises "Free use of SWE 1.7" on Pro, and the machine-readable model table still lists SWE-1.7 at zero input and output cost with a credit multiplier of 0 across tiers. We cannot tell whether the preview was extended or the prose is stale. If free access to Cognition's own model is part of why the price works for you, confirm it before subscribing.

We also did not verify the Cognition-reported figures at the time of the 2025 acquisition — $82 million ARR, 350+ enterprise customers, hundreds of thousands of daily active users. Those are the company's own claims, they are now thirteen months old, and they describe Windsurf before the founders left.

How much does Windsurf cost?

Frequently asked questions