GugleRAG v0.2.0-dev.15
Pre-release
Pre-release
GugleRAG v0.2.0-dev.15
English
- Added persistent, vendor-neutral document embeddings with cosine vector retrieval.
- Added SiliconFlow and local OpenAI-compatible embedding providers, with the complete default endpoint
https://api.siliconflow.cn/v1/embeddings. - Added optional SiliconFlow, local, and custom HTTP reranking providers.
- Added automatic startup and first-search indexing for documents created before this version, plus content-hash invalidation after edits.
- Added independent, expiring MCP access tokens with workspace scopes, listing, and revocation.
- Changed copied MCP configurations to the
httptype so clients can sendAuthorizationheaders. - Updated the setup and administrator retrieval settings, integration coverage, and retrieval documentation.
简体中文
- 添加数据库中立的持久化文档嵌入与余弦相似度向量检索。
- 接入 SiliconFlow 与本地 OpenAI 兼容嵌入服务，默认完整调用地址为
https://api.siliconflow.cn/v1/embeddings。 - 添加可选的 SiliconFlow、本地和自定义 HTTP 重排服务。
- 启动或首次搜索时自动为升级前创建的文档建立索引，文档编辑后按内容哈希自动重建向量。
- 添加独立、可过期的 MCP 访问令牌，支持工作区范围、列表查看和弃用。
- 将复制出的 MCP 配置类型改为
http，以便客户端发送Authorization请求头。 - 更新初始化向导、管理员检索配置、集成测试与检索文档。
Downloads / 下载
All archives are unsigned portable builds. Extract the archive, edit .env or use the setup wizard, then run the executable from the extracted directory so it can serve frontend/dist.
所有压缩包均为未签名便携构建。解压后编辑 .env 或使用初始化向导，并从解压目录运行可执行文件，以便服务端读取 frontend/dist。