在工具输出、日志、文件和 RAG 片段进入 LLM 前先做压缩，编码 Agent 省约 20% token，JSON 省 60-95%，答案不变。提供库、代理和 MCP 服务器三种用法。
What's Changed Port agentic RAG to GO by @qinling0210 in https://github.com/infiniflow/ragflow/pull/19500 Revert "fix(ingestion): allow re-parsing a document whose parse
…ire by @jay77721 in https://github.com/infiniflow/ragflow/pull/20358 refactor: use correct ragflow.conf as nginx config file by @JinHai-CN in https://github.com/infiniflo
… by @anistark in https://github.com/vibrantlabsai/ragas/pull/2496 feat: migrate DataCompyScore to collections API by @anistark in https://github.com/vibrantlabsai/ragas/p
… @nkch1k in https://github.com/explodinggradients/ragas/pull/2391 Migrate SummaryScore by @rhlbhatnagar in https://github.com/explodinggradients/ragas/pull/2376 feat: add
🎉 RAGLite v1.0: DuckDB, Qwen3, parallel insertion, benchmarking, better retrieval quality Release Highlights - 🐤 support for DuckDB ( 137) - 🐻 support for Qwen3 ( 124)
Unify Efficient Fine-tuning of RAG Retrieval, including Embedding, ColBERT,Cross Encoder