Neon Memory FAQs

Question 1

Does Neon Memory require cloud services or an internet connection?

Accepted Answer

No, Neon Memory is designed for offline operation with SQLite-native storage and has no cloud dependencies. While it supports cloud-based embedding providers like OpenAI for vector search, it also offers local Ollama integration or can run solely on FTS5 full-text search without any API keys or internet connection.

Question 2

What is the Adversarial Cleanup Pipeline?

Accepted Answer

This is an optional, LLM-driven 5-phase process (Consolidator, Adversary, Judge) that intelligently debates the fate of low-importance memories. It proposes to prune, promote, or merge entries, preventing valuable information from being deleted solely based on low access counts or simple score thresholds.

Question 3

What are the storage and deployment options for Neon Memory?

Accepted Answer

Neon Memory utilizes SQLite, FTS5, and sqlite-vec for persistent, efficient, and offline memory management without cloud dependencies. It can be deployed as an MCP server for Claude Code, integrated as a Node.js library, or used as a standalone CLI tool.

Question 4

What is Neon Memory and its primary purpose?

Accepted Answer

Neon Memory is a persistent memory layer for AI agents and Claude Code setups. It aims to improve AI retrieval quality by scientifically modeling memory decay using Ebbinghaus principles and intelligently managing memories through an optional LLM-driven adversarial cleanup pipeline.

Question 5

How does Neon Memory model 'forgetting' scientifically?

Accepted Answer

It implements Ebbinghaus temporal decay, a scientific model where memories fade predictably over time without reinforcement. Every recall of a memory doubles its stability, ensuring the system prioritizes relevant, reinforced knowledge and improves retrieval quality over time, similar to spaced repetition in human learning.

Neon Memory

Neon Memory

主要功能

使用案例

主要功能

使用案例