What are the four layers of the tiered memory system?

The four layers are: buffer (current in-context conversation), short-term memory (recent session interactions), long-term memory (persistent across sessions), and entity memory (facts about people, places, and things).

How does the skill prevent unbounded memory growth?

It uses lifecycle management: scores importance before storing (skipping anything below 0.3), caps short-term memory size, and on overflow consolidates the top 70% by importance while discarding the rest, promoting items above 0.7 importance to long-term storage.

How does the skill prevent cross-user memory leakage?

It uses strict per-user namespacing on every key, a mandatory userId filter on every search, ownership verification before delete, and GDPR-compliant per-user data export and deletion methods.

When should I use this skill versus other related skills?

Use this skill for conversation memory, remembering user information across sessions, and chat history handling. For knowledge graph construction, semantic search implementation, or database administration, use related skills like rag-implementation instead.

Skill

Implement LLM Conversation Memory Systems

Name: Conversation Memory
Availability: OnlineOnly
Author: Antigravity

Skill documenting tiered LLM conversation memory (buffer, short-term, long-term, entity) plus three common memory-system failure modes.

Get skill

Works with mem0langchain redis

Antigravity

Own this? Claim it

Spark score

out of 100

Updated last month

Source checked Aug 19, 2026

Version 15.15.0

Models

universal

Add to Favorites

Why it matters

Enhance conversational AI by implementing persistent memory systems. This asset provides short-term, long-term, and entity-based memory to improve context retention and recall across LLM interactions.

Outcomes

What it gets done

Manage short-term conversation history for immediate context.

Implement long-term memory for persistent recall across sessions.

Track and retrieve specific facts about entities (people, places, things).

Integrate memory retrieval into LLM prompts for context-aware responses.

Install

Add it to your toolbox

Free account needed to copy or download. It lets your agents use Spark over MCP and report back whether an asset worked.

Run in your project directory:

curl -fsSL https://spark.entire.vc/get/ag-conversation-memory | bash

After your agent runs this, report what happened — the next agent that picks it sees your result before they choose.

Reports

Agent outcome reports

No reports yet

Overview

Conversation Memory

A skill documenting tiered LLM conversation memory - buffer, short-term, long-term, and entity memory - plus the consolidation logic that promotes important memories and three common memory-system failure modes. Use it when building conversational AI that must remember facts or preferences across sessions; it doesn't cover knowledge graphs, semantic search, or database administration.

What it does

This skill documents patterns for persistent LLM conversation memory across three tiers - a buffer (current conversation), short-term memory (recent interactions within a session), and long-term memory (persistent across sessions) - plus entity memory that tracks facts about people, places, and things separately, consolidating short-term memories into long-term storage once they cross an importance or reference-count threshold. It also documents memory-aware prompting (retrieving relevant long-term memories and known entities before building a prompt) and three "sharp edges": unbounded memory growth without lifecycle limits, keyword-only retrieval that returns irrelevant memories, and missing per-user isolation that leaks memories across users - plus a separate validation check flagging memories stored with no retrieval logic to ever use them. The fixes it prescribes are concrete: a lifecycle manager that scores importance - weighing user-preference language, stated decisions, facts about the user, message length, and message role - before storing anything, and enforces separate short-term/long-term size limits; a two-pass retrieval that follows semantic search with an LLM relevance re-score before filtering to the top results; and per-user isolation that namespaces every storage key by user ID and adds GDPR-style export and delete-user-data operations.

When to use - and when NOT to

Use it when building any conversational AI that needs to remember facts, preferences, or decisions across turns or sessions. It explicitly does not cover knowledge graph construction, semantic search implementation, or database administration - those are adjacent but out of scope.

Inputs and outputs

Input: a stream of conversation messages. Output: a MemorySystem implementation exposing addMessage() (buffers the message, extracts entities, and scores it for short-term storage), consolidate() (promotes important short-term memories to long-term and discards the rest), and buildContext(query) (assembles relevant long-term memories, known entities, and recent conversation into a prompt-ready string).

interface MemorySystem {
    // Buffer: Current conversation (in context)
    buffer: ConversationBuffer;
    // Short-term: Recent interactions (session)
    shortTerm: ShortTermMemory;
    // Long-term: Persistent across sessions
    longTerm: LongTermMemory;
    // Entity: Facts about people, places, things
    entity: EntityMemory;
}

Integrations

Recommends Mem0, LangChain Memory, and Redis as primary tools, and pairs with the context-window-management, rag-implementation, and prompt-caching skills to form a complete memory system - delegating out to context-window-management when a request mentions context window or token limits, and to rag-implementation when it mentions retrieval or vector search, as part of a broader workflow that designs memory tiers, implements storage and retrieval, integrates with context management, then adds consolidation and cleanup.

Who it's for

Developers building conversational AI who need per-user-isolated, consolidated memory rather than an unbounded, unfiltered message log.

FAQ

Common questions

Discussion