llmem

Store and retrieve LLM memories in SQLite with semantic search.

Updated Apr 27, 2026

Search Tags:#embeddings #retrieval #memory #sqlite #semantic-search #dream #llmem

Ready to run

npx skills add https://github.com/MichielDean/LLMem --skill llmem

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: llmem
Source: https://github.com/MichielDean/LLMem/tree/main/skills/llmem
Command: npx skills add https://github.com/MichielDean/LLMem --skill llmem

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

LLMem provides a structured, SQLite-backed memory store with semantic search and relation traversal to empower LLM agents to remember, recall, and relate experiences over time.

Core Features & Use Cases

Structured memory types (fact, event, decision, and more), full-text search, semantic retrieval via embeddings, and a memory-relations graph.
Relationship traversal across memories to surface related context and enable complex recall.
Dream cycle for maintenance, decay, boosts, and merging near-duplicates to keep memories fresh.

Quick Start

Start by installing and running the llmem CLI, then add memories and use search with relation traversal to surface contextual memories.

Frequently Asked Questions about llmem

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I add persistent memory with semantic search to my LLM agents?▼

You can add persistent memory to LLM agents by storing structured memory types like facts and events in a SQLite-backed store, then retrieving them using semantic search via embeddings for contextual recall.

What is a structured memory store with relation traversal for LLMs?▼

A structured memory store for LLMs saves experiences as typed memories and links them in a relations graph, enabling agents to traverse connections and surface related context during complex recall tasks.

How do I set up SQLite-backed semantic search for AI memory retrieval?▼

You set up semantic search for AI memory by installing the memory CLI, configuring SQLite for storage, and enabling optional embeddings via Ollama with a local vector index to retrieve relevant memories.

Does semantic memory retrieval require an embeddings server like Ollama?▼

Embeddings via Ollama are optional for semantic memory retrieval, but enabling them with a local vector index is required to perform fast semantic search rather than standard full-text search across stored memories.

How does a dream cycle maintain long-term memory in LLM agents?▼

A background dream cycle maintains long-term memory by running maintenance routines that apply decay to old memories, boost important ones, and merge near-duplicates to keep the memory store fresh.

Can I use full-text search instead of semantic search for LLM memory?▼

You can use full-text search instead of semantic search for LLM memory if you do not configure the optional embeddings server, though you will lose the ability to retrieve memories based on semantic similarity.