Respan是一个由Y Combinator冬季班(YC W24)孵化的开源AI网关项目,采用插件化架构设计1。该项目允许开发者通过组装不同的插件来构建所需的AI基础设施1。Respan项目采用MIT或Apache-2.0许可证开源1,目前处于pre-alpha阶段1。
Respan团队创建这一项目源于在使用LiteLLM过程中遇到的可维护性问题,其包装代码曾超过10万行1。通过完全重新设计,团队实现了小核心加插件生态的架构模式,使得单个提供商的实现代码大幅精简,以Mistral provider为例仅需22行代码1。LiteLLM自2024年以来已为数千个组织提供服务1。
项目提供开源和商业化两种模式。开源项目地址为https://github.com/respanai/respan-gateway[1](#source-1),商业托管平台信息可在https://www.respan.ai/pricing获取[1](#source-1)。
Respan, a project incubated through Y Combinator's Winter 2024 batch, has launched as an open-source Rust AI gateway designed around a plugin architecture 1. The platform enables developers to assemble different plugins to construct customized AI infrastructure according to their specific needs 1.
The project emerged from the team's experience with maintainability challenges encountered while using LiteLLM, where wrapper code had grown to exceed 100,000 lines 1. Through a complete redesign, Respan implements a lean core combined with a plugin ecosystem approach 1. The new architecture demonstrates significant efficiency gains—the Mistral provider implementation requires only 22 lines of code 1.
Respan is currently in pre-alpha stage and is available under MIT or Apache-2.0 licensing 1. In addition to the open-source project, the team offers a commercial managed platform for organizations seeking hosted solutions 1.
评论
还没有评论,欢迎留下第一条。