Home

Dense-Mem

Dense-Mem gives your AI tools a shared long-term memory.

Most AI assistants remember only the current chat. Dense-Mem runs as your own memory service, so tools like Claude Code, Codex, or other MCP clients can store important facts, recall them later, and keep memory separate from any one chat window.

You can use Dense-Mem when you want an assistant to remember:

• your project names, goals, and current priorities
• preferences such as coding style, naming rules, and review standards
• decisions already made in earlier sessions
• important setup details for a repo or team
• facts that should survive after the chat ends

Dense-Mem is not an AI agent by itself. It is the memory system your AI tools call when they need to save or recall information.

Common Use Cases

Use case	Example
Personal coding memory	"Remember that I prefer small pull requests and explicit error handling."
Project memory	"Recall the deployment steps for my API service."
Team memory	"Remember the frontend team uses Bun and strict TypeScript."
AI assistant continuity	"Before answering, recall what you know about this project."
Safer corrections	"Actually, keep the new preference and replace the old one."

How It Works In One Minute

flowchart LR
  User[You] --> Assistant[AI assistant]
  Assistant --> DenseMem[Dense-Mem]
  DenseMem --> Store[(Memory store)]
  Store --> DenseMem
  DenseMem --> Assistant
  Assistant --> User

Loading

1. You tell an AI assistant something worth remembering.
2. The assistant sends that memory to Dense-Mem.
3. Dense-Mem stores the original evidence and checks whether it should become an active fact.
4. Later, the assistant asks Dense-Mem what it remembers.
5. Dense-Mem returns relevant memory for the assistant to use.

Start Here

Goal	Page
Try Dense-Mem before installing anything	Hosted Demo
Run Dense-Mem with the fewest choices	Quick Start
Learn what to do after it is running	Using Dense-Mem
Connect Claude Code, Codex, or another MCP client	MCP Clients
Understand memory conflicts and corrections	Conflicts And Corrections
Export or import reusable knowledge	Skill Packs
Manage teams, profiles, and API keys	Portals And API Keys
Enable Prometheus metrics and dashboards	Telemetry
Publish Dense-Mem with HTTPS or Redis	Public HTTPS And Redis
Review every important setting	Configuration
Understand the system design	Architecture
Read tool, API, and operations notes	Technical Reference

What You Need

For the easiest setup, you need:

• a computer or server that can run Docker
• an AI provider API key for embeddings
• one command line window
• one MCP client such as Claude Code or Codex

The quick start keeps everything local by default. You can make Dense-Mem public later, after the local version works.

What Dense-Mem Stores

Dense-Mem stores evidence first. An evidence fragment is the text that supports a memory. From that evidence, Dense-Mem can create typed claims and promote safe, non-conflicting claims into active facts.

That means memory is not only a pile of notes. It keeps track of where a memory came from and whether it became an active fact.

What Dense-Mem Does Not Do

Dense-Mem does not:

• replace your AI assistant
• choose truth silently when memories conflict
• browse the internet for facts
• manage your whole application
• expose the control portal as a public admin site by default

If Dense-Mem sees a comparable conflict, the assistant should ask you which memory to keep.