windsurf capabilities

Cascade can deploy web applications directly from the IDE to a public Netlify-backed URL with automatic rebuilds on subsequent deployments. Supported frameworks include Next.js, React, Vue, Svelte, and static HTML/CSS/JS.

Cascade is Windsurf's built-in agentic coding assistant that operates in three modes — Code (creates and modifies files, runs commands, installs dependencies), Plan (explores the codebase and produces a structured Markdown plan before acting), and Ask (read-only Q&A). It tracks developer actions in real time (flow awareness) to infer intent without requiring re-prompting.

Cascade Hooks execute custom shell commands at 12 named lifecycle events in the agent's workflow, enabling logging, security enforcement, quality checks, and team standardization. Pre-hooks can block agent actions by exiting with code 2.

Windsurf indexes the local codebase using a RAG approach (M-Query retrieval) to give Cascade and the autocomplete engine full awareness of the project without fine-tuning. Fast Context retrieves relevant snippets at query time, including from files not currently open.

Devin is a cloud-based autonomous software engineering agent that runs on its own virtual machine with desktop and browser access, continuing to work after the editor is closed. It is paired with Devin Local, an on-machine agent harness (currently in preview) that is intended to eventually replace Cascade.

Windsurf ships both a standalone VS Code-forked editor (the primary surface) and plugins for existing IDEs including JetBrains, Visual Studio Code, Visual Studio, Vim/Neovim, Jupyter Notebook, Google Colab (via Chrome extension), and Eclipse.

Cascade supports the Model Context Protocol (MCP), allowing users to connect external tools and services — such as GitHub, Slack, PostgreSQL, Brave Search, and Filesystem — directly into the agent's tool-call loop. MCP servers extend Cascade's capabilities beyond its built-in tools.

Memories are auto-generated contextual snapshots that Cascade creates during conversations to persist one-off facts across sessions. Rules are explicitly defined behavioral guidelines (coding conventions, constraints) that govern how Cascade behaves. AGENTS.md files provide directory-scoped instructions as a simpler alternative to Rules.

Windsurf Cascade offers 100+ AI models from multiple providers — including Windsurf's own SWE series, Anthropic Claude (Opus/Sonnet/Haiku), OpenAI GPT series, Google Gemini, DeepSeek, Kimi, and Grok — selectable per conversation. Free models (SWE-1.5, SWE-1.6) consume no credits; all other models are gated by a per-token credit multiplier.

Windsurf offers five tiers targeting individuals and teams. Free provides access to Cascade with free models only; Pro and Max are usage-based individual plans introduced March 2026; Teams and Enterprise add centralized billing, SSO/SCIM, admin controls, and remote indexing.

Windsurf Tab is an AI-powered code suggestion engine powered by a custom in-house model. Supercomplete mode displays multi-line diff suggestions (additions and deletions) near the cursor, while Autocomplete mode provides traditional at-cursor completions. It draws context from code, terminal activity, Cascade chat history, prior editor actions, and optionally the clipboard.

Cascade can generate shell commands from natural language (via Cmd/Ctrl+I), execute them in the integrated terminal, and reference terminal output or stack traces in conversations. Four levels of auto-execution automation are available, from fully manual to fully autonomous.

Cascade can perform live web searches and fetch specific URLs to retrieve up-to-date documentation, answers, and external context during agentic tasks. Web search can be triggered automatically by Cascade's intent detection or manually via the @web mention.

Workflows are Markdown-based, user-defined automation scripts that encode multi-step sequences of instructions for Cascade. They are invoked manually via slash commands (e.g. /pr-review) and can compose other workflows.