memory-manager_skill

Overview

This skill manages elizaOS agent memory, conversation history, and context windows to keep agents responsive and relevant. It provides pruning strategies, context-window tuning, and tools for archiving and consolidating important facts. The goal is to prevent context overload while preserving high-value long-term information.

How this skill works

The skill inspects short-term, long-term, and knowledge memory types and applies configurable operations: create, retrieve, search, update, and delete. It runs pruning routines (time-based and size-based), applies decay models for temporal relevance, and ranks memories by importance and recency. It also exposes monitoring hooks and embedding-based semantic search for efficient retrieval.

When to use it

• When conversation context grows beyond model limits
• Before sending prompts to ensure only relevant memories are included
• To archive or consolidate long-lived user facts
• When memory store costs or latency increase
• To enforce data retention or compliance policies

Best practices

• Define clear short-term vs long-term memory policies and limits
• Score memories for importance and last-access to guide pruning
• Use time-based decay to lower relevance of old transient items
• Batch memory operations to reduce I/O and cost
• Index and embed memories for semantic search and fast retrieval
• Monitor memory growth and automate scheduled pruning

Example use cases

• Trim a game-dev chat history to the most relevant 50 items before generating assets
• Automatically archive completed design conversations into long-term storage with importance tags
• Prune low-importance memories older than 30 days to control storage and context window size
• Consolidate repeated user preferences into a single high-importance long-term memory
• Run periodic audits that report memory count, average age, and top-k important memories