ReAct Agent Memory —Persistent Reasoning Context Across Sessions

I will build a ReAct agent with persistent memory — integrating long-term memory storage into the reasoning loop so the agent can recall relevant past observations, decisions, and outcomes when they are relevant to the current task, without flooding the context window with irrelevant historical content. ReAct agents without memory repeat investigative steps that a session-persistent agent would skip — re-querying information already retrieved, re-reasoning through decisions already made — wasting tokens and time on redundant work that memory-augmented agents avoid by retrieving and injecting only relevant past context.

The build covers a memory store (vector database or structured storage), relevance-based memory retrieval integrated before the thought step, memory write after each completed task, context window budget management for retrieved memories, and a memory inspection tool for debugging.