Built in Production | 200+ Sessions Indexed

Claude Total Memory

Memory Infrastructure for Claude Code

Built on Claude's native Auto Memory. Adds a search engine, a knowledge graph, and a full session pipeline on top of the notepad Anthropic gave you.

Learn More Contact Us
Scroll to explore
claude-total-memory
$ claude
> Remember that our API uses OAuth2 with PKCE flow
✓ Saved to memory
--- Next session ---
> How does our API handle authentication?
Claude: Based on your previous sessions, your API uses OAuth2 with
PKCE flow. This was established on March 3rd and confirmed in your
session on March 10th when you updated the token refresh logic.

Works With

Claude Code MCP Protocol Python Neo4j Docker
The Problem

Claude Forgets.
Every. Single. Time.

✗
Zero persistent context

Every new conversation starts completely blank. Claude has no memory of what you built, decided, or discussed yesterday.

✗
Lost decisions & context

Architectural decisions, debugging findings, and project conventions disappear the moment a session ends.

✗
Constant re-explanation

Teams waste hours re-explaining project structure, conventions, and history to Claude at the start of every session.

user@dev:~$ claude "continue the auth refactor"
 
Claude: I don't have context about a previous
auth refactor. Could you explain what you've
been working on?
 
# ...again. This is the 4th time this week.
 
user@dev:~$ claude "remember the JWT decision?"
 
Claude: I don't have access to previous
conversations. Each session starts fresh.
The Solution

Memory Infrastructure, Built for Production

Anthropic built the notepad. Claude Total Memory builds the database, the search engine, and the knowledge graph on top of it. Every session builds on everything that came before.

⚡
RAG-Powered Search

Semantic search across all past sessions with sub-200ms retrieval. Claude finds the exact context it needs, when it needs it.

🕸
Knowledge Graph

Entities, relationships, and decisions tracked across 200+ sessions. Over 5,000 entities and 6,000 relationship edges. Fully indexed.

⚙
Automated Session Management

One command closes a session, indexes all content, updates the knowledge graph, and commits everything. Zero manual steps.

🔌
MCP Integration

Native Model Context Protocol server. Memory tools are available directly inside Claude Code. No context-switching required.

🖥
Multi-Environment Support

Works on Linux, macOS, and WSL2. Fully portable with one-command setup that configures everything automatically.

🏢
Enterprise Ready

Per-project memory isolation, team deployment support, and deterministic workflows designed for production engineering environments.

How It Works

Three Steps to Full Memory Infrastructure

Claude Code already ships with basic memory. Claude Total Memory adds the search layer, the knowledge graph, and the automated session pipeline on top. No complex configuration. No workflow changes.

1

Install

One command sets up everything. Claude guides you through environment detection and tier selection. RAG indexing, knowledge graph, and MCP server configured automatically.

./setup.sh
2

Work

Use Claude Code normally. Memory captures context, decisions, and code patterns in the background. Nothing changes in your workflow.

claude "continue where we left off"
3

Remember

Next session, Claude searches 200+ past sessions in under 200ms. Every decision, every pattern, instantly available. Full architectural context from day one.

/close-session
Engineering Depth

Built Through Production, Not Theory

18 architectural decisions across 4 phases of development. Every design choice earned through real-world usage.

1
Phase 1: Monolith
Pre-December 2025
  • Single agent, 37-step pipeline
  • One Haiku agent handling everything sequentially
2
Phase 2: Decomposition
December 2025 to January 2026
  • Individual session files replace monolithic logs
  • 7-agent orchestrated design introduced
  • Delegated analysis pattern v2.0
3
Phase 3: Simplification
February 2026
  • Two parallel Haiku agents (simpler topology)
  • Shared title and summary extractor
  • Python rewrite v3.7.0 with deterministic workflow
4
Phase 4: Production Hardening
March 2026
  • Graphiti agent self-sufficient
  • Compact-proof design for crash recovery
  • v3.8.0 zero-bash main: spawns 2 agents, runs zero bash commands
  • Benchmark: 99.3/100 quality score
Infrastructure

What's Under the Hood

A complete infrastructure stack, purpose-built for the Claude ecosystem.

Claude Haiku / Haiku 4.5
Primary execution model
Claude Sonnet 4.5
Multi-agent coordination
BM25 + Vector Search
Hybrid search engine
Python 3
Implementation language
Neo4j
Knowledge graph storage
Graphiti
Entity and relationship extraction
MCP Protocol
Native Claude Code integration
Docker
Infrastructure dependencies
Included Agents

Specialized Agents, Built In

Claude Total Memory ships with specialized agents that handle context retrieval, session management, code quality, and security.

🔍
Context Researcher

Searches across memory, sessions, and the knowledge graph to gather relevant context. Returns verified summaries with source citations using a 4-phase anti-hallucination pipeline.

⚡
Session Closer

Two Haiku agents orchestrate the 11-step session closure pipeline: markdown conversion, AI summary, RAG indexing, knowledge graph extraction, quality checks, and git commit.

✓
Quality and Review

Meta-reviewer validates work quality. Synergy-checker ensures cross-file consistency. Fixer closes identified gaps. All work together in an automated review cycle.

🛡
Analysis and Security

Code analyzer maps dependencies and complexity. Security reviewer performs OWASP Top 10 analysis and secrets scanning. Debugging specialist handles root cause analysis.

See It In Action

See It In Action

Built in Production

Numbers That Speak for Themselves

Every metric below comes from real, production use. Not benchmarks run for a demo.

200+ Session Files Indexed
5,000+ Knowledge Graph Entities
6,000+ Relationship Edges
99.3/100 Quality Score
<200ms Search Retrieval
3 Deployment Tiers
About

Built by Business Wings LLC

Designed and developed by Aleksa Mirkovic. Claude Total Memory started as a Maps of Content system in December 2025 -- two months before Anthropic shipped native memory. When Anthropic released Auto Memory (MEMORY.md) in February 2026, Aleksa migrated to it and built the full infrastructure stack on top: session pipeline, RAG search, and knowledge graph.

✉ contact@businesswings.me 🌐 businesswings.me
Aleksa Mirkovic
Aleksa Mirkovic
Founder, Business Wings LLC

"I started solving this in December 2025 with an Obsidian MOC system -- two months before Anthropic shipped native memory. When they released Auto Memory, I migrated to it and built the full infrastructure layer on top: search, knowledge graph, session pipeline. Anthropic built the notepad. I built everything else."