Guides

Practical guides to agent memory.

These are not release notes. They are the questions people ask when an AI coding agent keeps forgetting decisions, preferences, project context, and the fixes they already paid for in time and tokens.

01

Agent memory security: how memory gets poisoned and what to do about it

Memory poisoning plants attacker-controlled text into an agent's memory. Remnic defends the write path with trust zones, provenance, and an explicit review queue.

02

How to give Cursor persistent memory

Point Cursor at a local Remnic server over MCP and it can recall and store project memory that survives new chats, restarts, and branch switches.

03

Export your ChatGPT memory and import it into Remnic

Request an OpenAI data export, then move your ChatGPT saved memories into Remnic with the @remnic/import-chatgpt package, a dry run, and provenance tagging.

04

What is procedural memory for AI agents?

Procedural memory stores how an agent does a recurring multi-step task, not just isolated facts. Remnic captures runbooks and mines them from real work.

05

Self-hosting an MCP memory server

Run Remnic as a standalone HTTP and MCP memory server: install the CLI, start the daemon, add bearer tokens and namespaces, or deploy the Docker image with QMD bundled.

06

One memory across Claude Code and Codex CLI

Install both connectors against one Remnic daemon and a preference you set in Claude Code can be recalled in Codex CLI. Memory stays scoped per project.

07

Import Supermemory memories into Remnic

Move a Supermemory JSON export into Remnic with the optional @remnic/import-supermemory package, dry-run checks, and local provenance.

08

Agent memory vs RAG: what is the difference?

Agent memory and RAG both retrieve context, but agent memory stores evolving facts, decisions, preferences, and procedures from work over time.

09

What is AI agent memory?

AI agent memory is the system that lets agents keep useful facts, decisions, preferences, and procedures across sessions.

10

How to give Claude Code persistent memory

Claude Code memory works best when recall happens before the model sees the task and durable facts are stored locally after the session.

11

Why coding agents forget context between sessions

AI coding agents forget context because session history, context windows, and tool calls are not durable project memory.

12

Local-first AI memory vs hosted memory APIs

Local-first AI memory keeps agent context on your machine, while hosted memory APIs trade control for managed infrastructure.

13

What is an MCP memory server?

An MCP memory server gives AI agents a standard way to store, recall, inspect, and share memory across compatible clients.