NeuronLayer

Turn any AI coding assistant into an intelligent one.

The Problem

AI coding assistants are powerful but forgetful:

| Problem | Stat | |---------|------| | Productivity paradox | 19% SLOWER with AI | | "Almost right" frustration | 66% of developers | | Trust issues | Only 29% trust AI code | | Context collapse | AI forgets mid-session | | Technical debt | 1.64x higher errors | | Code duplication | 4x more cloning |

The Solution

NeuronLayer gives AI assistants persistent, intelligent memory:

• Never forget context - Decisions, patterns, and history persist across sessions
• Self-documenting codebase - Architecture docs generate automatically
• Context rot prevention - AI stays sharp even in long sessions
• Pattern enforcement - Stops code duplication before it happens
• Test-aware suggestions - AI respects your tests

Compatibility

| Tool | Supported | Auto-Configure | |------|-----------|----------------| | memcode (Built-in agent) | Yes | N/A (Native) | | Claude Desktop | Yes | neuronlayer init | | Claude Code (CLI) | Yes | neuronlayer init | | OpenCode | Yes | neuronlayer init | | VS Code + Continue | Yes | Manual config | | Cursor | Not yet | No MCP support | | Any MCP Client | Yes | Manual config |

Quick Start

Installation

npm install -g neuronlayer

One-Command Setup

cd your-project
neuronlayer init

That's it! This automatically:

1. Registers your project
2. Configures Claude Desktop
3. Configures OpenCode
4. Configures Claude Code

Just restart your AI tool and NeuronLayer is active.

Output

NeuronLayer initialized!

Project: my-project
Path: /home/user/my-project
Data: ~/.neuronlayer/projects/my-project-abc123

Configured MCP Clients:
  ✓ Claude Desktop: ~/.config/claude/claude_desktop_config.json
  ✓ OpenCode: ~/.opencode/config.json
  ✓ Claude Code: ~/.claude.json

Restart your AI tools to activate.

Manual Configuration

If you prefer manual setup or use a different MCP client:

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "neuronlayer": {
      "command": "npx",
      "args": ["-y", "neuronlayer", "--project", "/path/to/your/project"]
    }
  }
}

Config locations:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Linux: ~/.config/claude/claude_desktop_config.json

OpenCode

Add to ~/.opencode/config.json:

{
  "mcpServers": {
    "neuronlayer": {
      "command": "npx",
      "args": ["-y", "neuronlayer", "--project", "/path/to/your/project"]
    }
  }
}

Features

Core Features

| Feature | Description | |---------|-------------| | Semantic Search | Find code by meaning, not just keywords | | Decision Recording | Log architectural decisions with context | | Pattern Library | Learn and enforce coding patterns | | File Indexing | Symbol extraction for TS/JS/Python |

Advanced Features

| Feature | Description | |---------|-------------| | Living Documentation | Auto-generated architecture docs | | Context Rot Prevention | Smart compaction keeps AI focused | | Confidence Scoring | Know when AI is guessing vs confident | | Change Intelligence | "What changed?" and "Why did it break?" | | Architecture Enforcement | Suggest existing functions, prevent duplication | | Test-Aware Suggestions | Predict test failures before they happen | | Super Intelligent Brain (Ghost Mode) | Proactively checks for conflicts, déjà vu, and session context | | Pre-commit Quality Gate | Validates AI-generated code against hallucination and patterns BEFORE commit |

How It Works

+-------------------------------------------------------------+
|                      NEURONLAYER                             |
+-------------------------------------------------------------+
|                                                              |
|  +--------------+    +--------------+    +--------------+   |
|  |   AI Tool    |--->|  MCP Server  |--->|   Memory     |   |
|  | Claude/Open  |    |   (stdio)    |    |   Engine     |   |
|  +--------------+    +--------------+    +--------------+   |
|                                                    |         |
|                            +--------------------+--+----+   |
|                            |                    v       |   |
|  +--------------+    +-----+----+    +------------------+|  |
|  |  Your Code   |--->| Indexer  |--->|  SQLite + Vec DB ||  |
|  |  (watched)   |    |  (Regex) |    |  (embeddings)    ||  |
|  +--------------+    +----------+    +------------------+|  |
|                                                          |  |
+-------------------------------------------------------------+

Memory Tiers

1. Tier 1 - Hot cache for active files (instant)
2. Tier 2 - SQLite for decisions, patterns, history (fast)
3. Tier 3 - Vector embeddings for semantic search (smart)

MCP Tools (The Gateway Architecture)

NeuronLayer elegantly compresses 51+ granular internal functions into 12 optimized Gateway Tools, dramatically saving prompt tokens and improving LLM routing reliability (proven 759x faster than standard grep).

6 Core Gateways

• memory_query - Semantic search, symbol lookup, and focused context retrieval.
• memory_record - Save architectural decisions, patterns, and feedback.
• memory_review - Pre-flight check before code changes vs patterns & decisions.
• memory_status - Project overview, architecture diagrams, and health checks.
• memory_ghost - The "Super Intelligent Brain": flags conflicts, déjà vu, and session resurrection.
• memory_verify - Pre-commit quality gate for AI-generated code (hallucinations, imports, security, tests).

6 Standalone Commands

• switch_project - Hot-swap active project context.
• switch_feature_context - Resume tracking an ongoing feature.
• trigger_compaction - Smart compression of context tree.
• update_decision_status - Mark architectural choices as deprecated/superseded.
• export_decisions_to_adr - Generate markdown Architecture Decision Records (ADRs).
• discover_projects - Find eligible Git repositories on disk.

🤖 memcode - The First-Party AI Coding Assistant

NeuronLayer powers its own fully-featured terminal coding agent: memcode. It natively understands NeuronLayer's context graph and avoids context rot.

Quick Start

# Set your preferred API key
export ANTHROPIC_API_KEY=your-key

# Run memcode inside your project
npx neuronlayer memcode

memcode Features

• Multi-LLM Support: Claude (Anthropic), all GPT/o1 (OpenAI), 200+ models via OpenRouter, and Local LLMs (Ollama).
• Persistent Memory: Decisions and patterns automatically persist across your chat sessions.
• Full Shell & Git access: Runs commands, tests, and edits files directly in your sandbox.
• Session Lifecycles: Save and resume conversations exactly where you left them.

Interactive Commands

Inside a memcode session, use:

• /model [name] - Hot-swap LLMs.
• /feature [name] - Track a specific feature area.
• /cost - Track precise token usage and cost.
• /save & /continue - Session management.

CLI Commands

# Quick setup (auto-configures Claude Desktop, OpenCode)
neuronlayer init

# Run the native AI Coding Assistant
neuronlayer memcode

# List all registered projects
neuronlayer projects list

# Add a new project
neuronlayer projects add /path/to/my-project

# Switch active project
neuronlayer projects switch <id>

# Export decisions to ADR files
neuronlayer export --format madr

# Discover projects in common locations
neuronlayer projects discover

# Show help
neuronlayer help

Data Storage

NeuronLayer stores project data separately for each project:

~/.neuronlayer/
├── projects/
│   ├── project-a-abc123/
│   │   ├── neuronlayer.db    # SQLite database
│   │   └── embeddings/       # Vector index
│   └── project-b-def456/
│       ├── neuronlayer.db
│       └── embeddings/
└── registry.json             # Project registry

Each project is isolated - no data mixing between projects.

Privacy

NeuronLayer is 100% local by default:

• All data stored on your machine
• No cloud services required
• No telemetry or tracking
• Works completely offline

Development

Prerequisites

• Node.js 18+
• npm or yarn

Setup

# Clone the repository
git clone https://github.com/abhisavakar/neuronlayer.git
cd neuronlayer

# Install dependencies
npm install

# Build
npm run build

# Run tests
npm test

# Type check
npm run typecheck

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

Areas We Need Help

• Additional language support (Go, Rust, Java)
• IDE extensions (VS Code, JetBrains)
• Cursor integration (when they add MCP support)
• Documentation improvements
• Performance optimizations

Roadmap

• [x] Core MCP server
• [x] Living Documentation
• [x] Context Rot Prevention
• [x] Confidence Scoring
• [x] Change Intelligence
• [x] Architecture Enforcement
• [x] Test-Aware Suggestions
• [x] Auto-setup for Claude Desktop
• [x] Auto-setup for OpenCode
• [ ] VS Code extension
• [ ] Cursor support (pending MCP)
• [ ] Team features (shared memory)

License

MIT License - see LICENSE for details.

Author

Abhishek Arun Savakar - savakar.com

Acknowledgments

Built with:

• Model Context Protocol by Anthropic
• better-sqlite3
• Xenova/transformers for embeddings

Made with determination over a weekend.

Code intelligence that transforms AI assistants into intelligent coding partners.