Skip to content

Atlantic83/NeoCoder

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Matrix Typing
 ███╗   ██╗███████╗ ██████╗    ██████╗ ██████╗ ██████╗ ███████╗██████╗ 
 ████╗  ██║██╔════╝██╔═══██╗  ██╔════╝██╔═══██╗██╔══██╗██╔════╝██╔══██╗
 ██╔██╗ ██║█████╗  ██║   ██║  ██║     ██║   ██║██║  ██║█████╗  ██████╔╝
 ██║╚██╗██║██╔══╝  ██║   ██║  ██║     ██║   ██║██║  ██║██╔══╝  ██╔══██╗
 ██║ ╚████║███████╗╚██████╔╝  ╚██████╗╚██████╔╝██████╔╝███████╗██║  ██║
 ╚═╝  ╚═══╝╚══════╝ ╚═════╝    ╚═════╝ ╚═════╝ ╚═════╝ ╚══════╝╚═╝  ╚═╝

NEOCODER

Python Claude License MCP

** AI-Native Engineering Agent for Enterprise-Scale Codebase**

InstallationQuick StartFeaturesWorkflow ModeCode ReviewConfigurationMCP ServerLiveBench Leaderboard


teaser.jpg


LiveBench Performance Radar Chart

⚡ Efficiency Comparison in real world COMPLEX task:

Efficiency Comparison Chart

🤖 SWE-Compass Benchmark

SWE-Compass

SWE COMPASS Leaderboard

🤖 AI/RUN Engineering Benchmark

Source: EPAM AI/RUN Engineering Benchmark

AI/RUN Engineering Benchmark Leaderboard

✨ Key Features

Feature Description
🧠 Orchestrator Core agent loop with iterative reasoning and action execution
🔧 Extensions Modular tool system (bash, file editing, code search, file_restore)
💾 Memory Management multi-layer memory architecture
🎯 Multiple Modes Coding, debugging, and refactoring modes
🚀 Workflow Preset 8-stage professional engineering pipeline with brainstorming, planning & testing
Auto Code Review Checks anti-patterns, bottlenecks, best practices, pitfalls & security on save
🖥️ Cross-Platform Full support for Windows, Linux, and macOS
🔌 MCP Integration Use as MCP server in Claude Code

🔌 Supported Providers

neoCoder supports multiple ways to connect to LLM APIs.

Anthropic

Direct connection to the official Anthropic API.

  • Uses the ANTHROPIC_API_KEY environment variable
  • Suitable for working with Claude models through the official endpoint

ZenMux

Connection through ZenMux — a unified API gateway that provides access to multiple LLM providers.

  • Uses the ZENMUX_API_KEY environment variable
  • Provides unified access to multiple LLM providers
  • Acts as the default provider

OpenAI

You can also use your own compatible API endpoint.

Google Vertex AI

You can also use your own compatible API endpoint.


🧠 Memory System

neoCoder uses a multi-layer memory architecture instead of a flat chat history.

Layers

  • Working Memory — active session context for the current task
  • Adaptive Memory — dynamically tunes compression threshold and rolling window based on task type
  • Context Compression — summarizes older turns when token pressure grows
  • Hierarchical Memory — persistent cross-session storage of task results and artifacts
  • Episodic Memory — stores completed tasks as reusable execution episodes
  • Procedural Memory — learns reusable workflows and successful patterns
  • Notes Storage — project notes and detailed markdown knowledge
  • xMemory — hierarchical semantic memory with facts and themes
  • Session Context — persists tech stack, brainstorming, plan, and progress in .neoCoder/context.md

memory.jpg

How it works

  1. On task start, neoCoder can retrieve relevant episodes, procedures, notes, and semantic knowledge.
  2. During execution, it keeps active context in working memory.
  3. When the context grows too large, it compresses older messages into structured summaries.
  4. After completion, it stores outcomes in persistent memory and creates reusable episodes.

This architecture makes neoCoder better at:

  • long-horizon engineering tasks,
  • cross-session continuity,
  • token-efficient context management,
  • reusing prior project experience,
  • structured planning and workflow persistence.

📦 Installation

Note

Requires Python 3.10+ and pipx or uv for the recommended global CLI install.

Recommended: pipx

pipx install neocoder

Alternative: uv

uv tool install neocoder

Local development

# Clone and install
cd neoCoder
pip install -e .

# With development dependencies
pip install -e ".[dev]"

# With MCP server support
pip install -e ".[mcp]"

Tip

neoCoder loads configuration from %USERPROFILE%\\.neoCoder\\settings.json on Windows and ~/.neoCoder/settings.json on Linux/macOS, so the installed CLI works from any directory.


🚀 Quick Start

Tip

Quick start in 3 steps: install → set API key → run

1️⃣ Initialize Configuration

neocoder \init-config          # Create example configuration

2️⃣ Set API Key

Important

API key from Anthropic or ZenMux is required

🪟 Windows CMD
set ANTHROPIC_API_KEY=your-key
set ZENMUX_API_KEY=your-key
💠 Windows PowerShell
$env:ANTHROPIC_API_KEY="your-key"
$env:ZENMUX_API_KEY="your-key"
🐧 Linux / macOS
export ANTHROPIC_API_KEY=your-key
export ZENMUX_API_KEY=your-key

3️⃣ Run

# Interactive mode
neocoder

# Single task
neocoder "Create a Python hello world script"

# With options
neocoder --mode debug "Fix test_api.py"
neocoder --preset thorough "Refactor auth module"

💻 Usage

Basic Commands

Command Description
neocoder Interactive mode
neocoder "task" Execute single task
neocoder --help Show help
neocoder \help Show system commands
neocoder \version Show version

✨ Professional Engineering Mode (Workflow Preset) ✨

The workflow preset is a production-grade, 8-stage development process with brainstorming, detailed planning, testing, and code review - designed for complex engineering tasks.

neocoder --preset workflow "your complex task"

8-Stage Engineering Pipeline:

Stage Description Auto-Approve
1️⃣ Problem Breakdown Decomposes task into subtasks with dependencies ❌ Requires approval
2️⃣ Brainstorming High-level approach, alternatives, trade-offs, risks ❌ Requires approval
3️⃣ Implementation Plan Creates detailed step-by-step plan based on brainstorming ❌ Requires approval
4️⃣ Implementation Executes the plan with file operations ✅ Auto-proceeds
5️⃣ Integration Testing Runs automated tests with retry logic ✅ Auto-proceeds
6️⃣ Tech Stack Review Validates against best practices ✅ Auto-proceeds
7️⃣ Code Review Automated review using project rules ✅ Auto-proceeds
8️⃣ Merge Preparation Generates commit-ready summary ✅ Auto-proceeds

Key Features:

  • 🔄 Stateful execution - Resume interrupted workflows
  • 📊 Extended context - 16K tokens, 250 iterations
  • 🎯 Deterministic - Temperature 0.0 for consistency
  • 🔍 Quality gates - Automatic testing and code review
  • 📝 Conventional commits - Auto-generated commit messages

Example Usage:

# Start a complex feature
neocoder --preset workflow "Add OAuth2 authentication with JWT tokens"

# Resume if interrupted (state is saved)
neocoder --preset workflow "Add OAuth2 authentication with JWT tokens"
# Prompt: Continue from previous state? (y/n)

# Reset workflow state
neocoder workflow reset

Real Complex Example

neocoder "Design a full-stack web application for a cryptocurrency trading bot dashboard.
The application should allow users to monitor their bot's performance in real-time, including current holdings, profit/loss, open trades, and historical performance data visualized through interactive charts.
Users should be able to configure trading parameters, such as risk tolerance, stop-loss levels, and take-profit targets, through an intuitive interface.
The application should feature secure authentication and authorization, protecting user data and trading credentials.
The backend should be built using Node.js with Express.js, utilizing a PostgreSQL database for data persistence.
The frontend should be built using React, with a clean and modern design emphasizing data visualization and ease of use.
The application should include robust error handling and logging capabilities.
The deployment should be considered, suggesting a suitable cloud platform and deployment strategy.
Provide the complete codebase, including frontend, backend, and database schema, ready for deployment.
The application should be named 'LovableBotDashboard'"

When to use Workflow Preset:

  • ✅ Complex features requiring multiple files
  • ✅ Refactoring with extensive testing
  • ✅ Production-critical changes
  • ✅ Team projects with code review requirements
  • ❌ Quick fixes or single-file changes (use default mode)

Standard Presets

# Fast mode - Quick iterations
neocoder --preset fast "Fix the login bug"

# Thorough mode - Extended analysis
neocoder --preset thorough "Refactor auth module"

Workflow Commands

neocoder --preset workflow "your task"    # Run with workflow preset
neocoder workflow status                   # Check workflow status
neocoder workflow reset                    # Reset workflow state

System Commands

💡 Use backslash \ for system commands

neocoder \init-config    # Create configuration
neocoder \help           # Show system commands  
neocoder \version        # Show version

🛡️ File Change Confirmation

Use --confirm (-c) flag to require manual approval before each file modification:

neocoder --confirm "create a new module"
neocoder -c "refactor auth system"

Confirmation dialog options:

Key Action
y Accept changes
n Reject changes
e Edit in external editor
d View full diff
a Accept all remaining changes
q Abort operation

🎨 Command Palette (Interactive Mode)

In interactive mode, use :: commands for quick actions:

Command Action
:: Open command palette with fuzzy search
::m Switch mode (coding/debug/refactor)
::t Switch or create thread
::s Toggle mode (coding ↔ debug)

Built-in commands:

Command Action
exit / quit / q Exit agent
reset Reset current session
help Show help


⚙️ Configuration

Configuration files are stored in ~/.neoCoder/:

~/.neoCoder/
├── 📄 settings.json        # Agent configuration
├── 📄 models.json          # Model definitions
├── 📄 tool_selection.json  # Tool selection settings
└── 📁 notes/               # Persistent notes from sessions

Environment Variables

Variable Description
ANTHROPIC_API_KEY Official Anthropic API key
ZENMUX_API_KEY ZenMux API key (default)
NEOCODER_BASE_URL Custom API endpoint
NEOCODER_CONFIG Custom config path

Review Priority Order

Priority Check Description
1️⃣ Anti-Patterns God Objects, Spaghetti Code, Shotgun Surgery
2️⃣ Bottlenecks N+1 queries, sync I/O in loops, inefficient operations
3️⃣ Best Practices Missing docs/JSDoc, mutable defaults, proper comparisons
4️⃣ Pitfalls Broad exception catching, empty catch blocks, assert misuse
5️⃣ Security Deserialization, command injection, path traversal, insecure HTTP

Additional Checks (all universal)

  • 🔒 Security: Hardcoded secrets, SQL injection, eval/exec, insecure YAML, debug flags
  • 🐛 Syntax: Parse errors (Python AST), brace/bracket balance (JS/TS)
  • 📊 Complexity: Cyclomatic complexity, nesting depth
  • 🎨 Style: Code duplication, dead code / unused imports
  • 🏷️ Types: Missing type hints (Python), any usage (TypeScript)

Example Output:

✓ Code Review: Score 85/100, 2 issue(s)
[critical] No Anti-Patterns (line 45): God Object: Class 'UserManager' has 25 methods (>20)
  → Fix: Split into smaller, focused classes (Single Responsibility Principle)

Configuration: Code review rules can be customized in .neoCoder/code_review_config.json


🔌 MCP Integration

Warning

MCP server requires separate installation: pip install -e ".[mcp]"

Run neoCoder as an MCP server for Claude Code:

neocoder-mcp
# or
python -m neoCoder.mcp.server

Claude Code Configuration

Add to your Claude Code MCP settings (claude_desktop_config.json or via claude mcp add):

{
  "mcpServers": {
    "neocoder": {
      "command": "python",
      "args": [
        "-m",
        "neoCoder.mcp.server"
      ],
      "env": {}
    }
  }
}

Or using the CLI:

claude mcp add neocoder -- python -m neoCoder.mcp.server

With uv (recommended):

claude mcp add neocoder -- uv run --directory /path/to/neoCoder python -m neoCoder.mcp.server

🏗️ Architecture

arch.jpg




**Made with ❤️ for developers**


About

AI-Native Engineering Agent for Enterprise-Scale Codebase

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors