CWindsurf

KO | EN

gitlite — search

#cpp #machine-learning #cli #tailwindcss #java #bash #android #video #http #claude-code

(★ 9)

Windsurf++ 让你在 Windsurf/Devin 中使用自己的模型 API Key、本地推理服务和兼容网关，并直接通过 Cascade 原生模型选择器切换模型。

File Explorer

ZIP Download

- icon.svg
README.md
README.zh-CN.md

# Use via CDN

jsDelivr

jsDelivr serves any public GitHub repository as a CDN with zero setup. Pick a version and a file to get a ready-to-paste link and snippet.

Version

File

Link

Example

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

> Bring Your Own Key for Windsurf and Devin.

[English](README.md) | [简体中文](README.zh-CN.md)

Windsurf++ lets you use your own model API keys, local inference servers, and compatible gateways in Windsurf/Devin while keeping model selection inside Cascade's native model picker.

It consists of a cross-platform installer CLI, the Windsurf++ extension, and a local proxy runtime. Official traffic remains independent when a request does not target a custom model or when BYOK authorization is unavailable.

> [!IMPORTANT]
> Windsurf++ is not an official product of Windsurf, Devin, Codeium, or any model provider. Installation modifies third-party application files and may be affected by application updates and applicable terms. Use it only on devices, accounts, and APIs you are authorized to access. Do not use it to bypass billing, access controls, or service restrictions.

## Quick Start

Node.js 18 or later and an installed copy of Windsurf or Devin are required.

```bash
# Install
npm install -g @priders/cwindsurf
cwindsurf install

# Check status
cwindsurf status

# Uninstall
cwindsurf uninstall
```

After installation:

1. Fully quit and reopen Windsurf/Devin.
2. Open **Windsurf++** from the activity bar.
3. Sign in with LinuxDO or GitHub to authorize the device.
4. Add a provider with its API endpoint, API key, and model definitions.
5. Test the connection and enable BYOK.
6. Select the custom model from Cascade's native model picker.

## Features

- **Native model selection** — Custom models appear directly in Cascade's model list.
- **Multiple protocols** — OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, and Google Gemini APIs.
- **Local and cloud models** — Ollama, LM Studio, vLLM, llama.cpp, and common compatible cloud services.
- **Agent tool calling** — Streaming text, reasoning/thinking, function calls, tool-result continuation, and multi-turn context.
- **Visual configuration** — Manage providers, models, context limits, output limits, and protocol options from the sidebar.
- **Stream reliability** — Idle timeouts, exponential backoff, rate-limit delays, and controlled reconnect behavior.
- **Usage and cost analytics** — Tokens, estimated cost, latency, throughput, success rate, cache usage, and budget alerts.
- **Local commit messages** — Choose a provider, model, and language for commit-message generation.
- **Device authorization** — LinuxDO/GitHub sign-in, device management, remote revocation, and signed leases.
- **Cross-platform installer** — macOS, Windows, and Linux.

## Provider Protocols

| Type | Standard endpoint | Typical use |
|---|---|---|
| `openai` | `/chat/completions` | Ollama, LM Studio, vLLM, DeepSeek, OpenRouter, and compatible services |
| `openai-responses` | `/responses` | OpenAI Responses API and compatible gateways |
| `anthropic` | `/messages` | Anthropic Claude API and Messages-compatible services |
| `gemini` | `/models/*:streamGenerateContent` | Google Gemini `v1` / `v1beta` streaming API |

Compatibility depends on whether the target service fully implements streaming, tool calling, tool-result continuation, and usage fields. Run a connection test after adding a provider.

## How It Works

```text
Windsurf / Devin
  │
  ├─ Local interception and proxy layer
  │   ├─ Identify official requests and BYOK models
  │   └─ Keep official traffic on an independent pass-through path
  │
  └─ Windsurf++ Extension
      ├─ Provider and model management
      ├─ OpenAI / Responses / Anthropic / Gemini adapters
      ├─ Streaming and tool-call conversion
      └─ Local analytics, authorization, and runtime status
```

The installer creates backups before modifying application files. `cwindsurf uninstall` removes the patches and extension and can optionally back up user configuration first.

## Local Data and Privacy

Configuration is stored under `~/.cwindsurf/`. Provider API keys stay on the local device. Runtime statistics are aggregated locally by default, and full prompts or responses are not required for normal analytics.

Never publish API keys, OAuth tokens, refresh tokens, device private keys, release signing keys, or configuration files containing secrets.

## Platform Support

| Platform | Status |
|---|---|
| macOS Apple Silicon / Intel | Supported |
| Windows | Supported |
| Linux | Supported |

Application updates may overwrite patches. Run `cwindsurf status` after updating Windsurf/Devin and reinstall when necessary.

## Issues and Feedback

This repository is intended for project information, release notes, and issue tracking. The main project source code is not included.

When opening an issue, include your OS and architecture, application and Windsurf++ versions, provider protocol, model ID, reproduction steps, and sanitized error details.

## License

The project and distributed packages are currently marked `UNLICENSED`. Public availability does not grant permission to copy, modify, redistribute, or use the project commercially unless the maintainers provide an explicit license.

CWindsurf

Owner Repos

File Explorer

# Use via CDN

Was this content helpful?

Recent Feedback

CWindsurf

Owner Repos

File Explorer

# Use via CDN

Was this content helpful?

Recent Feedback

Download README