Vespper是一家Y Combinator F24孵化的初创公司，推出了一款模型上下文协议（MCP）产品，使AI代理能够高效编辑Word文档1。该产品通过微调模型将Word文档转换为HTML供AI进行编辑，随后将更改协调回原始.docx文件，相比现有替代方案在速度上快3倍、成本低2倍，且准确度更高1。
Vespper的MCP设计简洁，仅暴露读取、搜索和编辑三个工具，相比DOCX skill需要10次工具调用和Office CLI的13次调用，平均每个任务仅需3次调用1。该产品基于3-8B微调模型与LoRA适配器，专门用于将HTML差异协调回OOXML格式1。免费层用户每月可获得500次编辑次数，产品支持云端运行、零数据保留或自托管多种部署方式1。
Vespper的客户已涵盖法律科技公司、简历优化AI初创、政府科技公司和生命科学初创等多个领域1。公司创始人包括Topaz（前Snyk高级软件工程师）和Dudu（前Viz.ai深度学习工程师）1。
Vespper, a Y Combinator F24-backed startup, has introduced a Model Context Protocol (MCP) product that enables AI agents to efficiently edit Word documents 1. The tool converts .docx files to HTML for AI editing, then reconciles changes back to the original format using a fine-tuned 3-8B model with LoRA adapters specialized for OOXML conversion 1.
The platform significantly outperforms existing alternatives, operating three times faster and at half the cost while delivering higher accuracy compared to competing solutions 1. Vespper's MCP exposes only three core functions—read, search, and edit—requiring an average of just three tool calls per task, substantially fewer than the ten calls needed by DOCX Skill or thirteen by Office CLI 1. A free tier provides 500 monthly edits, with deployment options including cloud hosting, zero-data retention infrastructure, or self-hosted setups 1.
The company's customer base spans multiple sectors, including legal technology firms, resume optimization AI startups, government technology companies, and life sciences startups 1. Vespper was founded by Topaz, a former senior software engineer at Snyk, and Dudu, previously a deep learning engineer at Viz.ai 1.
评论
还没有评论，欢迎留下第一条。