Cascade, and what session awareness actually means
The central feature is an agent called Cascade, and its distinguishing property is memory of what you have been doing — files opened, edits made, commands run, decisions taken earlier in the session.
The practical effect is that instructions can refer backwards. “Now do the same for the other two providers” works without restating what “the same” was. “Undo that last approach and try it with a queue instead” is a complete instruction. Conventional chat assistants lose that thread between messages and require you to rebuild context each time.
That is a genuine difference in how it feels to work, and it is the reason people who like Windsurf like it specifically rather than generally. It is also difficult to evaluate in a ten-minute trial — the advantage appears over an hour, not a task.
Supercomplete, and the parts you still type
Alongside the agent sits inline prediction for the moments you are writing directly. It is competent and it is not the reason to choose this editor; if fast autocomplete is your main requirement, the Codeium plugin gives you that without changing editors.
The editor also indexes the codebase, so questions and agent tasks reach code you have not opened — the same mechanism as Cursor, with the same caveat that a stale index answers confidently about code that has changed.
The corporate history, which is not trivia
This tool’s ownership has moved twice in short order. Codeium the company renamed itself Windsurf, then — following a licensing arrangement involving Google — Windsurf was acquired by Cognition in 2025.
Day to day that changes nothing about using it. It matters for two decisions.
Standardising a team on it means betting on a product whose strategic direction now belongs to a different company than the one that built it, and whose relationship to Cognition’s other products is not fully settled.
Reading older reviews is unreliable — anything written before mid-2025 describes a different company, and pricing and plan structures have moved since. Verify current terms directly rather than trusting a comparison article, including this one.
Hosted by default, self-hosted by negotiation
On standard plans, code is processed by hosted models. There is no local model option at the consumer tiers.
Enterprise arrangements exist, including self-hosted and hybrid deployment options — which is unusual in this category and is a real differentiator against Cursor for organisations that need it. Those terms are negotiated rather than published, so treat any specific claim about them as something to confirm rather than assume.
What running it involves
- Windows, macOS or Linux. A desktop application.
- A free tier that is genuinely usable, which is rare among agentic editors and makes evaluation cheap.
- Paid plans meter agent usage; heavy sessions consume an allowance quickly, and running out mid-task is the common complaint.
- An internet connection at all times on standard plans.
- Willingness to change editors — VS Code settings and extensions import, but it is a different application that trails upstream.
Whether agent-first suits how you work
This is the decision, and it is a preference rather than a capability question.
Some developers genuinely want to delegate whole tasks: describe an outcome, go and do something else, come back and review a diff. For them, an editor arranged around that is better shaped than one where the agent is a panel bolted to the side.
Others think by writing, and want the tool to accelerate their hands rather than replace them. For those developers, Windsurf’s interface emphasises a mode they will use occasionally, and Cursor or Copilot fits better.
Neither preference is wrong and neither is a skill level. Try the free tier for a real task before deciding, because this is not something a feature list resolves.
Who should be using it
Developers who want to work at the level of delegated tasks, and who are comfortable reviewing results rather than steps.
People working in unfamiliar codebases, where indexing plus session awareness reduces how much you must explain before anything useful happens.
Teams needing an enterprise deployment story that Cursor does not offer.
It is a weaker fit for developers who want completion above all, for anyone unwilling to change editors, and for organisations that need stability of ownership as well as of software.
Its strengths
- Session continuity — the agent remembers the last hour, not just the last message.
- Multi-file execution with commands and self-correction from their output.
- Codebase indexing, so questions reach beyond open files.
- A usable free tier, more than most agentic editors offer.
- Enterprise deployment options, including self-hosting on negotiated terms.
- Familiar foundations — still VS Code underneath.
The reasons to hesitate
- Proprietary and hosted on standard plans, with no local models.
- Ownership changed twice in a short period; strategic direction is now Cognition’s.
- Agent-first is a real trade — if you prefer to drive, the interface emphasises a mode you will not use.
- Another editor to adopt, and a fork that trails upstream VS Code.
- Usage limits bite during heavy agent work, exactly when throttling hurts most.
- Stale indexes mislead with no indication they are stale.
Switching away
Your code is untouched and your VS Code settings came from VS Code. The index, chat history and Cascade’s session memory are not portable and there is no export.
Switching out costs about what switching in cost — an afternoon of reconfiguration. Nothing is stranded, which is the usual position for editor tools and worth contrasting with platform tools such as Replit.
The realistic comparisons
- Cursor IDE — the closest competitor, stronger on completion, similar agent, more settled ownership.
- Cline — agent capability in stock VS Code, with your own API keys and no editor change.
- Claude Code — terminal-based, editor-independent, one provider.
- Codeium — the same company’s plugin, for completion without changing editors.
Compiled from Windsurf’s documentation and public sources. Ownership and plan structures changed during 2025 — verify current terms directly. We have not hands-on tested this tool. Last reviewed 16 August 2026.
