In Memoria

In Memoria

MCP server that gives AI coding assistants persistent memory and pattern learning.

AI coding tools suffer from complete memory loss between sessions. Every conversation with Claude, Copilot, or Cursor starts from scratch, wasting time and tokens re-explaining your codebase.

In Memoria solves session amnesia by building persistent intelligence about your code that AI tools can access through the Model Context Protocol.

The Problem

# What happens now
You: "Refactor this function using our established patterns"
AI: "What patterns? I don't know your codebase."
You: *explains architecture for the 50th time*

# What should happen
You: "Refactor this function using our established patterns"
AI: "Based on your preference for functional composition and your naming conventions..."

Current AI tools:

• Re-analyze codebases every session (expensive)
• Give generic suggestions that don't match your style
• Have no memory of architectural decisions
• Can't learn from corrections you've made

Technical Approach

In Memoria runs as an MCP server that AI tools connect to. It provides 17 tools for codebase analysis and pattern learning.

Architecture:

┌─────────────────────┐    MCP    ┌──────────────────────┐    napi-rs    ┌─────────────────────┐
│  AI Tool (Claude)   │◄─────────►│  TypeScript Server   │◄─────────────►│     Rust Core       │
└─────────────────────┘           └──────────┬───────────┘               │  • AST Parser       │
                                             │                          │  • Pattern Learner  │
                                             │                          │  • Semantic Engine  │
                                             ▼                          └─────────────────────┘
                                   ┌──────────────────────┐
                                   │ SQLite + SurrealDB   │
                                   │  (Local Storage)     │
                                   └──────────────────────┘

Core engines:

• AST Parser (Rust): Tree-sitter based parsing with complexity analysis and symbol extraction
• Pattern Learner (Rust): Analyzes coding decisions and builds developer style profiles
• Semantic Engine (Rust): Maps code relationships and architectural concepts
• TypeScript Layer: MCP server, database operations, file watching
• Storage: SQLite for structured data, SurrealDB for vector operations and semantic search

Quick Start

# Start the MCP server
npx in-memoria server

# Connect from Claude Desktop (add to config)
{
  "mcpServers": {
    "in-memoria": {
      "command": "npx",
      "args": ["in-memoria", "server"]
    }
  }
}

# Connect from Claude Code CLI
claude mcp add in-memoria -- npx in-memoria server

What It Learns

Coding Patterns:

// Learns you prefer functional composition
const processUser = pipe(validateUser, enrichUserData, saveUser);

// Future suggestions match this style, not OOP alternatives

Naming Conventions:

// Learns your patterns: useXxxData for API hooks, handleXxx for events
const useUserData = () => { ... }
const handleSubmit = () => { ... }

// Suggests consistent naming in new code

Architecture Decisions:

// Remembers you use Result types for error handling
type ApiResult<T> =
  | { success: true; data: T }
  | { success: false; error: string };

// Suggests this pattern instead of throwing exceptions

MCP Tools

Analysis:

• analyze_codebase - Get architectural overview and metrics
• search_codebase - Semantic search (find by meaning, not keywords)
• get_file_content - Retrieve files with metadata
• get_project_structure - Project hierarchy understanding

Intelligence:

• get_semantic_insights - Code relationships and dependencies
• get_pattern_recommendations - Suggestions matching your style
• predict_coding_approach - How you'd solve similar problems
• get_developer_profile - Your learned preferences
• learn_codebase_intelligence - Trigger analysis

Automation:

• auto_learn_if_needed - Smart learning when gaps detected
• contribute_insights - AI can add back to knowledge base
• get_learning_status - What's been analyzed
• quick_setup - Project initialization

System:

• get_system_status - Health and diagnostics
• get_intelligence_metrics - Learning quality metrics
• get_performance_status - Performance benchmarks

Implementation Details

Pattern Learning Algorithm:

1. Parse code into ASTs using tree-sitter
2. Extract structural patterns (function signatures, class hierarchies, naming)
3. Build frequency maps of developer choices
4. Train classifier on decision patterns
5. Generate predictions for new code contexts

Performance:

• Incremental analysis (only processes changed files)
• SQLite for structured data, SurrealDB embedded for semantic search and vectors
• Cross-platform Rust binaries (Windows, macOS, Linux)
• Handles codebases up to 100k files

Team Usage

In Memoria works for individual developers and teams:

Individual:

• Learns your personal coding style
• Remembers architectural decisions you've made
• Provides context-aware suggestions

Team:

• Share intelligence across team members
• Onboard new developers with institutional knowledge
• Maintain consistent AI suggestions team-wide

# Export team knowledge
in-memoria export --format json > team-intelligence.json

# Import on another machine
in-memoria import team-intelligence.json

Technical Comparison

vs GitHub Copilot's memory:

• Copilot: Basic fact storage, no pattern learning
• In Memoria: Semantic analysis with prediction engine

vs Cursor's rules:

• Cursor: Static rules, manually defined
• In Memoria: Dynamic learning from actual code

vs Custom RAG:

• RAG: Retrieves relevant code snippets
• In Memoria: Understands coding patterns and predicts behavior

Build from Source

git clone https://github.com/pi22by7/in-memoria
cd in-memoria
npm install
npm run build

Requirements:

• Node.js 18+
• Rust 1.70+ (for building)
• 2GB RAM minimum

Development:

npm run dev          # Start in development mode
npm test            # Run test suite
npm run build:rust  # Build Rust components

Contributing

This is infrastructure for the AI development ecosystem. Contributions welcome:

• Language support (add tree-sitter parsers)
• Pattern learning improvements
• MCP tool additions
• Performance optimizations

See CONTRIBUTING.md for details.

FAQ

Q: Does this replace my AI coding assistant?A: No, it enhances them. In Memoria provides memory and context that AI tools can use.

Q: What data is collected?A: Everything stays local. No data is sent to external services.

Q: How accurate is pattern learning?A: Accuracy improves with codebase size. Help us benchmark this across different codebases to establish solid metrics.

Q: Performance impact?A: Minimal. Runs in background, incremental updates only.

License

MIT - see LICENSE

Try it: npx in-memoria server

Give your AI tools the memory they've been missing.