Launched in november 2024

Windsurf Wave 1

2024-12-11

1. Cascade memories (windsurfrules): Ability to specify rules for Cascade to follow via Cascade memories (windsurfrules)
2. Automated terminal commands : Ability to automate execution of certain commands without needing manual acceptance each time
3. WSL, devcontainer, Pyright support
4. Image input support (already launched)

Automated terminal commands

Automated terminal commands are meant to allow for safe execution of code autonomously, completing the loop of code modification and verification. When we first launched, all commands had to be approved by the developer before execution. In Wave 1, we have provided Cascade a list of commands that are always safe or have minimal side effects (e.g. ls, kubectl get), commands that are ambiguous (e.g. tar, pip, npm), and commands that are generally considered dangerous (e.g. rm). Cascade will automatically execute the safer commands without requiring developer intervention, while the rest still needs approval. Users have the ability to make modifications to these lists and whitelist/blacklist any commands of their choosing to control the level of autonomy that Cascade can have with regards to code execution. We even added an opt-in setting to let the AI decide whether a command is safe to run in the ambiguous cases, which will always prioritize obeying the allow/deny lists. See documentation for automated terminal commands here.

Cascade memories .windsurfrules

Cascade memories are meant to help Cascade know how to do work, not just what the end work output should look like. During training of LLMs, we usually just have the end state of the code, but anyone who has written code knows how the process is anything but linear or instant. Because AI flows like Cascade are meant to work lock-step with the developer, we want to guide Cascade, and this guidance is what we’re dubbing memories. This becomes even more important as the AI flow does work more autonomously. In Wave 1, a user can specify rules, which are the first interface for memories, and Cascade will obey them as it does work. This works similarly to rules or guidelines in other tools, and there are two sets of rules: global rules that are followed independent of workspace, specified within the Cascade UI, and workspace-specific local rules, which are specified in a .windsurfrules file. See documentation for Cascade memories here.

extensions

WSL support, devcontainer support, and Pyright for static type-checking for Python. This all comes on top of the features that we released prior to this Wave 1 announcement, such as image input support in Cascade and settings imports from Cursor.