CEMS

Continuous Evolving Memory System

Persistent memory for AI coding assistants. Your agent remembers what you teach it — across sessions, projects, and teams.

Works with Claude Code · Cursor · Codex · Goose · any MCP-compatible agent

---

How It Works

1. You code — CEMS hooks into your IDE (Claude Code, Cursor, Codex, Goose)
2. CEMS watches — Every prompt is enriched with relevant memories. Session learnings are extracted automatically.
3. Memories stored — Preferences, conventions, architecture decisions, debugging fixes
4. Context recalled — Next session, relevant memories are injected so your agent already knows your codebase

No manual tagging. No copying context. It just works.

---

Architecture

---

Quick Start

1. Deploy the server

# Download the compose file
curl -fsSLO https://raw.githubusercontent.com/chocksy/cems/main/deploy/docker-compose.yml
# Create .env
cat > .env << 'EOF'POSTGRES_PASSWORD=your_secure_passwordOPENROUTER_API_KEY=sk-or-your-keyCEMS_ADMIN_KEY=cems_admin_random_stringEOF# Start
docker compose up -d
# Run migrationsforfin migrate_docs_schema.sql migrate_soft_delete_feedback.sql migrate_conflicts.sql;do
curl -fsSL "https://raw.githubusercontent.com/chocksy/cems/main/scripts/$f"| \
docker exec -i cems-postgres psql -U cems cems
done

2. Create users

cems admin --admin-key $CEMS_ADMIN_KEY users create alice
# Returns the API key — save it, shown only once!

3. Install the client

Each developer runs:

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

Prompts for server URL and API key, then configures your IDE. Done.

---

Documentation

Doc	What's inside
Deployment Guide	Docker Compose, Kubernetes, env vars, backups, production checklist
Client Setup	Install options, CLI commands, skills, hooks, updating, troubleshooting
Architecture	Storage, search pipeline, maintenance, observer daemon, MCP
API Reference	All REST endpoints with examples

---

Development

git clone https://github.com/chocksy/cems.git &&cd cems
uv pip install -e ".[dev]"
pytest

License

MIT