langchain-short-memory

Manage short-term memory in LangChain by applying thread-scoped persistence, trimming, summarization, and selective deletion to bound conversation history. These lifecycle patterns keep model performance stable as multi-turn conversations grow.

Thread-scoped memory in LangGraph chat agents isolates conversation history using a thread_id, ensuring each multi-turn chat maintains its own bounded context. This enables per-thread persistence, trimming, and summarization without cross-conversation interference.

Summarize conversation history for LangChain agents by setting summarization triggers that activate when the message window exceeds a defined size. This condenses older interactions into a summary, retaining context relevance while controlling token usage.

Yes, you can selectively delete messages from LangGraph thread memory through safe deletion handling. This pattern removes specific messages from conversation history while maintaining the integrity of the remaining thread context.

The best way to trim LangChain chat agent memory is to apply message window sizing that bounds the number of retained messages per thread. This operational pattern drops older messages systematically, keeping recent context relevant for model performance.

You need to be using LangChain or LangGraph to implement thread-level memory persistence and deletion patterns. The operational requirements specify thread_id scope, message window sizing, and safe deletion handling applicable to these frameworks.