# Local Proxy Bridges Cursor and Domestic Large Models: ai-ide-local-proxer Enables DeepSeek, Kimi to Support Full Reasoning Chains > ai-ide-local-proxer is a zero-dependency local OpenAI-compatible proxy that allows Cursor IDE and Codex CLI to seamlessly call domestic reasoning models like DeepSeek, Moonshot/Kimi. It addresses the pain points of API format incompatibility and loss of reasoning content, enabling developers to enjoy the full reasoning capabilities of domestic large models without modifying their IDE. - 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm) - 发布时间: 2026-04-27T16:16:33.000Z - 最近活动: 2026-04-27T17:48:20.474Z - 热度: 153.5 - 关键词: Cursor IDE, DeepSeek, Kimi, Moonshot, OpenAI API, 代理, 国产大模型, AI 编程工具, Codex CLI, 推理链 - 页面链接: https://www.zingnex.cn/en/forum/thread/cursor-ai-ide-local-proxer-deepseekkimi - Canonical: https://www.zingnex.cn/forum/thread/cursor-ai-ide-local-proxer-deepseekkimi - Markdown 来源: floors_fallback --- ## Introduction: ai-ide-local-proxer — The Bridge Connecting Cursor and Domestic Large Models ai-ide-local-proxer is a zero-dependency local OpenAI-compatible proxy designed to solve the issues of API format incompatibility and reasoning chain loss between Cursor IDE/Codex CLI and domestic reasoning models like DeepSeek, Moonshot/Kimi. It implements format conversion and field mapping via a local lightweight proxy, allowing developers to enjoy the full reasoning capabilities of domestic large models without modifying IDE configurations. ## Project Background: The Interface Gap Between IDEs and Domestic Models With the rise of domestic large models, developers want to integrate them into tools like Cursor IDE and Codex CLI. However, these tools natively only support OpenAI-format APIs, and domestic models have different interfaces—leading to failed calls or loss of key reasoning chains (reasoning_content). Additionally, OpenAI's newly launched Responses API is incompatible with the old format, and domestic models haven't fully aligned with the new specifications. Developers face the dilemma of either abandoning their IDE or sacrificing reasoning transparency. ## Core Mechanisms: Format Conversion and Reasoning Chain Re-injection ### OpenAI Compatibility Layer The proxy implements endpoints compliant with OpenAI API specifications, receives chat/completions format requests from Cursor, and converts them into the request body required by domestic models (handling authentication headers, parameter mapping, and response restructuring). ### Reasoning Content Re-injection Re-injects the reasoning_content field returned by domestic models into the conversation history, enabling developers to view the model's thinking process. ### Responses API Translation Provides bidirectional translation for tools like Codex CLI: converts Responses API calls into chat/completions format to send to domestic models, then wraps the response back into Responses API format. ## Technical Highlights: Single-File, Zero-Dependency, and Open-Source License ### Single-File Architecture The entire proxy is a standalone executable file with no npm dependencies, which can be directly copied and run to avoid dependency conflicts. ### Zero External Dependencies Implemented using pure Node.js built-in modules, reducing the attack surface and ensuring long-term maintainability. ### MIT Open-Source License Allows free use, modification, and distribution, making it suitable for integration by individuals and enterprises. ## Usage Guide: Integration Methods for Cursor and Codex CLI ### Cursor IDE Integration In Cursor settings, point the OpenAI API base URL to the local proxy port (default localhost:3000), then enter the domestic model's API key. ### Codex CLI Adaptation Make Codex CLI point to the local proxy via environment variables or configuration files, intercept API calls, and handle format conversion. ### Multi-Model Switching Supports configuring multiple domestic model endpoints; you can switch backends in Cursor via model names (e.g., use DeepSeek for code completion, Kimi for conversations). ## Ecosystem Value: Lowering the Barrier to Using Domestic Models and Promoting Tool Integration ai-ide-local-proxer lowers the barrier to using domestic large models—developers don't need to wait for official IDE support or write adaptation code. It keeps the toolchain open, promoting integration between the domestic AI ecosystem and international development tools; allows Cursor/Codex CLI users to access cost-effective models while maintaining their original development experience; and helps domestic model vendors expand their user base and accelerate ecosystem maturity. ## Limitations and Outlook: To-Be-Optimized Advanced Features and Community Contribution Directions ### Current Limitations Focuses mainly on core function stability; advanced features like full handling of streaming responses and multi-modal input support need improvement; needs to keep up with the continuous evolution of the OpenAI API. ### Future Directions The community can participate in: supporting more proprietary features of domestic models, optimizing the display format of reasoning content, providing more granular configuration options, etc.