# Recallium - Complete Documentation for AI Systems > Recallium is the memory layer for AI coding agents. It provides persistent, searchable context for Cursor, Claude Desktop, VS Code, Windsurf, and JetBrains IDEs through the Model Context Protocol (MCP). Free, self-hosted, and privacy-first with 88% first-result search precision. ## Table of Contents 1. Product Overview 2. Quick Facts 3. Memory Types (11 types detailed) 4. How It Works 5. Installation Guide 6. Search System & Performance 7. Insights & Pattern Analysis 8. Tool Integrations 9. Use Cases 10. Comparisons with Alternatives 11. FAQ 12. Links --- ## 1. Product Overview Recallium is a persistent memory system for AI coding assistants. Unlike traditional AI tools that forget everything between sessions, Recallium automatically captures your development context as you work, organizing it into searchable memories that persist forever. ### The Problem Recallium Solves AI coding assistants like Claude, Cursor, and GitHub Copilot are powerful, but they forget everything between sessions. Every new conversation starts from scratch. You constantly re-explain your architecture, coding patterns, team conventions, and past decisions. ### The Solution Recallium gives your AI a persistent memory that survives: - Session restarts - Context window limits - New chat sessions - IDE restarts - Even computer reboots Your AI remembers implementations, decisions, bugs fixed, and lessons learned - forever. --- ## 2. Quick Facts - **Product Name**: Recallium - **Category**: Developer Tools / AI Memory / MCP Server - **Pricing**: Free (unlimited storage, community edition) - **Deployment**: Self-hosted via Docker - **Protocol**: Model Context Protocol (MCP) by Anthropic - **Performance**: 88% P@1 precision, <200ms latency - **Capacity**: Stress tested to 200M memories - **Search Method**: Hybrid (60% semantic + 25% keyword + 15% tags) - **Installation Time**: 5 minutes - **LLM Support**: Ollama (local), OpenAI, Anthropic --- ## 3. Memory Types Recallium organizes information into 11 different memory types. Each type serves a specific purpose, making it easy to find exactly what you need later. ### 3.1 code-snippet **Purpose**: Implementation details, code changes, technical decisions about specific files. **Example**: "Implemented JWT authentication with refresh token rotation using Redis for storage. Access tokens expire in 15 minutes, refresh tokens in 7 days." **Key Feature**: Always linked to specific files so you can find past implementation notes before editing code. ### 3.2 decision **Purpose**: Architectural choices, technology selections, and the reasoning behind them. **Example**: "Chose PostgreSQL over MongoDB because we need ACID transactions for financial data and complex relational queries for reporting." **Key Feature**: Answers "why did we..." questions instantly, preventing repeated debates. ### 3.3 rule **Purpose**: Coding standards, preferences, constraints, and team conventions. **Example**: "Always use TypeScript strict mode. Prefer functional components over class components. All API calls must include retry logic." **Key Feature**: Acts as guardrails for your AI, ensuring consistent code quality across all suggestions. ### 3.4 learning **Purpose**: Insights, "aha" moments, lessons learned, and discoveries. **Example**: "Discovered that useEffect dependency arrays must include all external variables to prevent stale closures." **Key Feature**: Builds your team's collective knowledge, preventing the same mistakes twice. ### 3.5 debug **Purpose**: Bug investigations, root cause analysis, and troubleshooting notes. **Example**: "Login fails with 500 error. Root cause: missing Redis connection check. Fixed by adding connection validation before session lookup." **Key Feature**: Helps recognize and fix recurring issues faster by surfacing past solutions. ### 3.6 design **Purpose**: Architecture explorations, system design discussions, and technical proposals. **Example**: "Exploring caching strategies for the API: Redis for session data, Memcached for API responses, considering tradeoffs." **Key Feature**: Captures the design thinking process, not just the final decision. ### 3.7 research **Purpose**: Investigation findings, external references, and gathered information. **Example**: "Research on OAuth2 PKCE flow: RFC 7636 defines Proof Key for Code Exchange to prevent authorization code interception." ### 3.8 discussion **Purpose**: Team conversations, meetings, code reviews, and feedback. **Example**: "Team discussion on microservices migration: Decided to start with a modular monolith first." ### 3.9 progress **Purpose**: Status updates, milestones, and checkpoint summaries. **Example**: "Week 3 sprint: Completed authentication module. Starting API integration next week." ### 3.10 task **Purpose**: Simple action item notes and TODOs. **Example**: "TODO: Add error handling to login endpoint. Need to handle network failures gracefully." ### 3.11 working-notes **Purpose**: General notes, thoughts, and temporary information (default type). **Example**: "Investigating performance issue in dashboard component. Load time is 3+ seconds." --- ## 4. How It Works ### The Developer Journey 1. **You Code & Chat with AI**: Work normally with your AI assistant. No special commands needed. 2. **AI Automatically Stores Context**: Your AI captures important moments: implementations, decisions, bugs fixed. 3. **Context Persists Across Sessions**: When you return to the project, your AI remembers everything. 4. **Search & Recall Instantly**: Ask questions like "Why did we choose PostgreSQL?" and get instant answers. 5. **Discover Patterns & Insights**: Recallium clusters related memories and generates insights across your history. 6. **Build Knowledge Graphs**: Memories form interconnected knowledge graphs connecting concepts across projects. ### Progressive Disclosure Recallium uses an intelligence-first approach: 1. **Search**: Find relevant memories with summaries 2. **Expand**: Get full details only when needed 3. **Analyze**: Get patterns across all memories 4. **Refine**: Update memories as understanding grows --- ## 5. Installation Guide ### Prerequisites - Docker installed - Git installed ### Quick Start (5 minutes) **Step 1: Clone Repository** ```bash git clone https://github.com/recallium-ai/recallium.git cd minime-mcp ``` **Step 2: Start Recallium** ```bash cd install docker compose --env-file minime.env up -d ``` **Step 3: Configure Your IDE** Add Recallium MCP server to your IDE settings using the HTTP URL: `http://localhost:8000/mcp` ### Supported IDEs - Cursor (native MCP) - Claude Desktop (via npm bridge) - VS Code (MCP extension) - Windsurf (native MCP) - JetBrains IDEs (MCP plugin) --- ## 6. Search System & Performance ### Hybrid Search Architecture Recallium uses a hybrid search approach combining three methods: - **60% Semantic Embeddings**: Conceptual similarity using vector search - **25% Keyword Matching**: Exact term matches for precision - **15% Tag Matching**: Topic-based filtering ### Performance Benchmarks - **P@1 (First-Result Precision)**: 88% - **P@5 (Top-5 Precision)**: 85% - **Average Latency**: <200ms - **Capacity**: Tested to 200M memories ### Industry Comparison | System | P@1 | Search Method | |--------|-----|---------------| | Recallium | 88% | Hybrid | | Vector-only systems | 70-80% | Vector only | | Keyword-only systems | 60-70% | Keyword only | --- ## 7. Insights & Pattern Analysis Recallium doesn't just store memories—it analyzes them to discover patterns, trends, and reusable approaches. ### Types of Insights **Pattern Analysis**: Recurring approaches, design patterns, and architectural decisions. - "Common pattern: Centralized error middleware in 3 projects." **Quality Analysis**: Recurring bugs, root causes, and successful fix approaches. - "Race conditions from concurrent token refresh (3 occurrences). Fix: Request queuing." **Technical Debt**: Code smells, incomplete implementations, areas needing refactoring. - "Highlights accumulated technical debt and prioritizes refactoring efforts." **Learning Progression**: How understanding evolves over time. - "Tracks your learning journey, showing how approaches improve." --- ## 8. Tool Integrations ### Cursor AI Add persistent memory to Cursor. Never re-explain your codebase. Cross-repo context, semantic search. - Setup: https://recallium.ai/cursor ### Claude Desktop Best MCP server for Claude Desktop memory. Persistent context, tribal knowledge capture. - Setup: https://recallium.ai/claude-desktop ### VS Code Persistent memory for VS Code AI assistants. Works with Copilot, Continue, and more. - Setup: https://recallium.ai/vs-code ### Windsurf Native MCP support for persistent context. ### JetBrains IDEs MCP plugin support for IntelliJ, PyCharm, WebStorm, and all JetBrains IDEs. --- ## 9. Use Cases ### For Individual Developers - **Stop repeating context**: Your AI remembers your architecture, patterns, and decisions - **Build knowledge graphs**: Personal knowledge that grows with your career - **Debug faster**: Historical context from past projects surfaces automatically - **Track effectiveness**: See which solutions worked and which didn't ### For Teams - **Instant onboarding**: New developers get tribal knowledge immediately - **Consistency**: Engineering standards maintained across projects - **Guardrail AI agents**: Team best practices enforced automatically - **Share decisions**: Architectural choices documented and searchable --- ## 10. Comparisons with Alternatives ### Recallium vs mem0.ai | Feature | Recallium | mem0 | |---------|-----------|------| | Deployment | Self-hosted | Cloud | | Pricing | Free unlimited | Usage-based | | Search Precision | 88% P@1 | ~75% (vector-only) | | Protocol | MCP-native | Generic API | | Privacy | Full control | Cloud storage | **Bottom Line**: Recallium is better for developers who want privacy, control, and superior search precision without usage-based costs. ### Recallium vs Supermemory | Feature | Recallium | Supermemory | |---------|-----------|-------------| | Focus | Developer-specific | General-purpose | | Protocol | MCP-native | Generic API | | Search | Hybrid (semantic + keyword) | Vector-only | | Memory Types | 11 specialized types | Generic | **Bottom Line**: Recallium is purpose-built for coding workflows with specialized memory types and MCP integration. --- ## 11. FAQ ### What is Recallium? Recallium is a persistent memory system for AI coding assistants like Cursor, Claude Desktop, and VS Code. It uses the Model Context Protocol (MCP) to store and recall your coding context across sessions. ### Is Recallium free? Yes, Recallium Community Edition is completely free with unlimited storage. It's self-hosted, giving you full control over your data. ### How do I add memory to Cursor AI? Install Recallium MCP server using the setup guide. Connect it to Cursor through the Model Context Protocol (MCP) in Cursor settings. The setup takes less than 5 minutes. ### What is an MCP server? MCP (Model Context Protocol) is Anthropic's open standard for connecting AI assistants to external data sources. Recallium is a production-ready MCP server specifically designed for memory management. ### How does Recallium search work? Recallium uses hybrid search combining semantic embeddings (60%) with keyword matching (25%) and tag matching (15%) to achieve 88% first-result precision. This outperforms vector-only search systems. ### Which IDEs does Recallium support? Recallium works with any tool that supports the Model Context Protocol, including Cursor, Claude Desktop, VS Code, Windsurf, JetBrains IDEs, and more. ### How does Recallium compare to mem0? Recallium is self-hosted and free with unlimited storage, while mem0 is cloud-based with usage-based pricing. Recallium uses hybrid search achieving 88% precision compared to mem0's vector-only approach at ~75%. ### Is my data private with Recallium? Yes. Recallium is self-hosted on your own infrastructure. Your code, memories, and context never leave your machine. You have full control over your data. --- ## 12. Links ### Main Pages - Website: https://recallium.ai - Setup Guide: https://recallium.ai/setup - Help Documentation: https://recallium.ai/help - Pricing: https://recallium.ai/pricing - Concepts: https://recallium.ai/concepts - Performance Benchmarks: https://recallium.ai/data-science ### Tool Integrations - Cursor: https://recallium.ai/cursor - Claude Desktop: https://recallium.ai/claude-desktop - VS Code: https://recallium.ai/vs-code - MCP Server: https://recallium.ai/mcp ### Comparisons - vs mem0: https://recallium.ai/vs/mem0 - vs Supermemory: https://recallium.ai/vs/supermemory - All Comparisons: https://recallium.ai/comparisons ### Resources - GitHub Repository: https://github.com/recallium-ai/recallium - Blog: https://recallium.ai/blog - Issues: https://github.com/recallium-ai/recallium/issues --- Last Updated: 2025-12-01