Basic Memory

Basic Memory lets you build persistent knowledge through natural conversations with Large Language Models (LLMs) like Claude, while keeping everything in simple markdown files on your computer. It uses the Model Context Protocol (MCP) to enable any compatible LLM to read and write to your local knowledge base.

What is Basic Memory?

Most people use LLMs like calculators - paste in some text, expect to get an answer back, repeat. Each conversation starts fresh, and any knowledge or context is lost. Some try to work around this by:

• Saving chat histories (but they're hard to reference)
• Copying and pasting previous conversations (messy and repetitive)
• Using RAG systems to query documents (complex and often cloud-based)

Basic Memory takes a different approach by letting both humans and LLMs read and write knowledge naturally using standard markdown files. This means:

• Your knowledge stays in files you control
• Both you and the LLM can read and write notes
• Context persists across conversations
• Context stays local and user controlled

How It Works in Practice

Let's say you're working on a new project and want to capture design decisions. Here's how it works:

1. Start by chatting normally:

We need to design a new auth system, some key features:

- local first, don't delegate users to third party system
- support multiple platforms via jwt
- want to keep it simple but secure

... continue conversation.

2. Ask Claude to help structure this knowledge:

"Lets write a note about the auth system design."

Claude creates a new markdown file on your system (which you can see instantly in Obsidian or your editor):

---
title: Auth System Design
permalink: auth-system-design
tags
- design
- auth
---

# Auth System Design

## Observations

- [requirement] Local-first authentication without third party delegation
- [tech] JWT-based auth for cross-platform support
- [principle] Balance simplicity with security

## Relations

- implements [[Security Requirements]]
- relates_to [[Platform Support]]
- referenced_by [[JWT Implementation]]

The note embeds semantic content (Observations) and links to other topics (Relations) via simple markdown formatting.

3. You can edit this file directly in your editor in real time:

# Auth System Design

## Observations

- [requirement] Local-first authentication without third party delegation
- [tech] JWT-based auth for cross-platform support
- [principle] Balance simplicity with security
- [decision] Will use bcrypt for password hashing # Added by you

## Relations

- implements [[Security Requirements]]
- relates_to [[Platform Support]]
- referenced_by [[JWT Implementation]]
- blocks [[User Service]]  # Added by you

4. In a new chat with Claude, you can reference this knowledge:

"Claude, look at memory://auth-system-design for context about our auth system"

Claude can now build rich context from the knowledge graph. For example:

Following relation 'implements [[Security Requirements]]':
- Found authentication best practices
- OWASP guidelines for JWT
- Rate limiting requirements

Following relation 'relates_to [[Platform Support]]':
- Mobile auth requirements 
- Browser security considerations
- JWT storage strategies

Each related document can lead to more context, building a rich semantic understanding of your knowledge base. All of this context comes from standard markdown files that both humans and LLMs can read and write.

Everything stays in local markdown files that you can:

• Edit in any text editor
• Version via git
• Back up normally
• Share when you want to

Technical Implementation

Under the hood, Basic Memory:

1. Stores everything in markdown files
2. Uses a SQLite database just for searching and indexing
3. Extracts semantic meaning from simple markdown patterns
4. Maintains a local knowledge graph from file content

The file format is just markdown with some simple markup:

Frontmatter

• title
• type
• permalink
• optional metadata

Observations

• facts about a topic

- [category] content #tag (optional context)

Relations

• links to other topics

- relation_type [[WikiLink]] (optional context)

Example:

---
title: Note tile
type: note
permalink: unique/stable/id  # Added automatically
tags
- tag1
- tag2
---

# Note Title

Regular markdown content...

## Observations

- [category] Structured knowledge #tag (optional context)
- [idea] Another observation

## Relations

- links_to [[Other Note]]
- implements [[Some Spec]]

Basic Memory will parse the markdown and derive the semantic relationships in the content. When you run basic-memory sync:

1. New and changed files are detected
2. Markdown patterns become semantic knowledge:

• [tech] becomes a categorized observation
• [[WikiLink]] creates a relation in the knowledge graph
• Tags and metadata are indexed for search

3. A SQLite database maintains these relationships for fast querying
4. Claude and other MCP-compatible LLMs can access this knowledge via memory:// URLs

This creates a two-way flow where:

• Humans write and edit markdown files
• LLMs read and write through the MCP protocol
• Sync keeps everything consistent
• All knowledge stays in local files.

Using with Claude

Basic Memory works with the Claude desktop app (https://claude.ai/):

1. Install Basic Memory locally:

{
  "mcpServers": {
    "basic-memory": {
      "command": "uvx",
      "args": [
        "basic-memory"
      ]
    }
}

2. Add to Claude Desktop:

Basic Memory is available with these tools:
- write_note() for creating/updating notes
- read_note() for loading notes
- build_context() to load notes via memory:// URLs
- recent_activity() to find recently updated information
- search() to search infomation in the knowledge base

3. Install via uv

uv add  basic-memory

# sync local knowledge updates
basic-memory sync

# run realtime sync process
basic-memory sync --watch

Design Philosophy

Basic Memory is built on some key ideas:

• Your knowledge should stay in files you control
• Both humans and AI should use natural formats
• Simple text patterns can capture rich meaning
• Local-first doesn't mean feature-poor

License

AGPL-3.0