跳转至

Node 04: Memory System (中文 + English)

导航 / Navigation: 返回项目首页 | 文档首页 | 中文 README | English README

Memory 是平台级节点 / Memory as a platform-level node

中文： Memory 不是聊天历史的别名，而是独立服务。当前支持 scope/type： - conversation - project - agent - execution - global

English: Memory is not just chat history. It is an independent service. Current scope/type values: - conversation - project - agent - execution - global

Memory APIs

  • POST /api/v1/memory
  • GET /api/v1/memory
  • POST /api/v1/memory/semantic-search
  • POST /api/v1/memory/retry-embeddings

自动向量化 / Automatic Embedding

中文： POST /api/v1/memory 默认 auto_embedding=true，服务端自动调用 Embedding Provider。

English: POST /api/v1/memory uses auto_embedding=true by default, so vectors are generated server-side.

降级策略 / Fallback Strategy

中文： - embedding 失败时，memory 记录仍会写入 - embedding_status=failed - 记录 embedding_error - 自动入重试队列，可异步重试

English: - If embedding generation fails, memory record is still saved - embedding_status=failed - embedding_error is stored - Retry job is queued and can be retried asynchronously

混合检索 / Hybrid Retrieval

中文： 语义检索支持 similarity_weight：

hybrid_score = similarity_weight * similarity_score + (1 - similarity_weight) * importance_score

English: Semantic retrieval supports similarity_weight:

hybrid_score = similarity_weight * similarity_score + (1 - similarity_weight) * importance_score

代码位置 / Code References

  • backend/app/api/v1/memory.py
  • backend/app/services/memory_store.py
  • backend/app/runtime/embeddings.py
  • backend/app/schemas/memory.py
  • backend/app/db/models.py