98% Recall@5 accuracy

Your AI assistant
finally has memory.

Stop re-explaining your codebase every session. CEMS gives Claude Code, Cursor, Codex, and Goose persistent memory — decisions, preferences, and context that actually carry over. 

$ curl -fsSL https://getcems.com/install.sh | bash

Works with your tools

Claude Code
Cursor
Codex
Goose
Any MCP Client
98%
Recall@5 accuracy
<50ms
Memory injection latency
100%
Self-hosted & private

How it works

Three steps to persistent memory. No learning curve.

1

Install and forget

One command hooks into your existing AI tools. No config files, no workflow changes.

2

It watches and learns

CEMS observes your sessions, extracts meaningful signals, and stores them privately.

3

Every new session starts informed

Relevant memories injected as context. No re-explaining solved problems.

Everything your AI needs to remember

A complete memory system, not just a vector store.

🧠

Persistent Memory

Decisions and preferences survive session boundaries. Your AI remembers what you decided last week.

👁️

Observer Daemon

Automatically extracts learnings from sessions. Runs in the background — zero effort from you.

🔌

MCP Server

Standards-compliant MCP server. Works with any compatible AI tool out of the box.

🛡️

Gate Rules

Block or warn on dangerous operations before they execute. Guardrails that learn from experience.

⚡

Profile Injection

Your preferences, coding style, and conventions injected at every session start.

🔒

Self-Hosted & Private

Your memories never leave your infrastructure. Full control over your data.

See it in action

Memory that surfaces exactly when you need it.

claude code session
$ claude "deploy the API to staging"
CEMS injected 3 memories:
#1 Staging deploy uses coolify app deploy api-staging
#2 Always run migrations before deploy: rake db:migrate
#3 Last deploy failed because Redis wasn't restarted first
Claude: Based on your previous deploys, I'll run migrations first and restart Redis before deploying to staging...

Frequently asked questions

Quick answers to common questions about CEMS.

Does CEMS work with my current tools?
CEMS integrates with Claude Code, Cursor, OpenAI Codex, and Goose out of the box. It also ships an MCP server that works with any MCP-compatible AI tool. One install command and you're done.
Where is my data stored?
CEMS is fully self-hosted. Your memories are stored in a PostgreSQL database on your own infrastructure. Nothing is sent to third-party services. You own your data completely.
How is this different from CLAUDE.md?
CLAUDE.md is a static file you manually maintain. CEMS is a dynamic memory system that automatically learns from your sessions, surfaces relevant context, and evolves over time. It also works across multiple AI tools, not just Claude.
Will it slow down my AI assistant?
Memory injection adds less than 50ms to session start. The observer daemon runs asynchronously and never blocks your AI assistant. You won't notice it's there.
What happens if I uninstall?
Your AI tools work exactly as they did before — they just lose their memory. Your data stays in your PostgreSQL database. Reinstall anytime to pick up where you left off.

Give your AI a memory that works.

Install in 60 seconds. Self-hosted. Free and open source. 

$ curl -fsSL https://getcems.com/install.sh | bash
View on GitHub | Documentation