Milvus: How to Turn Conversation History into Long-Term Memory
TL;DR · AI Summary
Milvus proposes a method to convert raw conversation history into readable, editable long-term memory using Markdown and semantic search.
Key Takeaways
- Store conversation history in Markdown format for readability and editability.
- memsearch is an open-source tool for agent memory management.
- Semantic and hybrid search improve contextual understanding for agents.
Outline
Jump quickly between sections.
Mindmap
See how the topics connect at a glance.
查看大纲文本（无障碍 / 无 JS 友好）
- Agent Memory Management
- 数据格式
- Markdown 存储
- 搜索技术
- 语义搜索
- 混合搜索
- 工具支持
- memsearch (GitHub)
Highlights
Key sentences worth saving and sharing.
The key is to make memory readable and editable as Markdown files.
Layer on semantic search and hybrid search so agents can recover the right context by meaning.
You can explore this workflow with memsearch, an open-source project we built for agent memory.
At last month’s Unstructured Data Meetup London, Jiang Chen, our Head of Developer Relations, broke down a question more agent builders are running into: 𝗵𝗼𝘄 𝗱𝗼 𝘆𝗼𝘂 𝘁𝘂𝗿𝗻 𝗿𝗮𝘄 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻 𝗵𝗶𝘀𝘁𝗼𝗿𝘆 𝗶𝗻𝘁𝗼 𝘂𝘀𝗲𝗳𝘂𝗹 𝗹𝗼𝗻𝗴-𝘁𝗲𝗿𝗺 𝗺𝗲𝗺𝗼𝗿𝘆? Raw conversation logs are a good starting point, but they should not become another black box. The key is to make memory readable and editable as Markdown files, then layer on semantic search and hybrid search so agents can recover the right context by meaning, even when no one remembers the exact keyword. You can explore this workflow with memsearch, an open-source project we built for agent memory. Full video: youtu.be/3mDFw933wdE?ut
memsearch on GitHub: github.com/zilliztech/mem#AgentMemory#VectorSearch#RAG
