AndyAI - Noundry AI Development Assistant

An AI-powered CLI tool for the Noundry platform that works like Claude Code - with conversational mode, tool execution, and MCP server integration. Supports both local models (CodeLlama) and cloud providers (Anthropic, OpenAI, Google, Nebius, and custom OpenAI-compatible APIs).

     _      _   _   ____   __   __       _      ___
    / \    | \ | | |  _ \  \ \ / /      / \    |_ _|
   / _ \   |  \| | | | | |  \ V /      / _ \    | |
  / ___ \  | |\  | | |_| |   | |      / ___ \   | |
 /_/   \_\ |_| \_| |____/    |_|     /_/   \_\ |___|

AndyAI - Noundry AI Development Assistant
Like Claude Code, but for Noundry

What AndyAI Does

AndyAI is an AI coding assistant that can:

• Chat conversationally about your code and project
• Read and write files in your project
• Execute shell commands and analyze output
• Search your codebase for files and patterns
• Generate code using Noundry patterns and TagHelpers
• Connect to MCP servers for extended tool capabilities

Two modes available:

• Local Mode: 100% private with CodeLlama - no API keys, no cloud
• Hosted Mode (BYOK): Use your own API keys for Anthropic, OpenAI, Google, Nebius, or any OpenAI-compatible endpoint

Quick Start

Option A: Use Hosted Providers (Recommended for best quality)

# Build
dotnet build

# Configure your API key (choose one)
dotnet run -- config set anthropic.apikey sk-ant-your-key
dotnet run -- config set openai.apikey sk-your-key
dotnet run -- config set nebius.apikey your-nebius-key

# Set active provider
dotnet run -- config set provider anthropic

# Start chatting
dotnet run -- chat

Option B: Run 100% Local

1. Download the Model (~4GB)

# Windows
.\scripts\download-model.ps1

# Linux/Mac
./scripts/download-model.sh

2. Build and Run

dotnet build
dotnet run -- config set mode Local
dotnet run -- chat

3. Start Chatting

> Create a contact form with validation

Andy: I'll create a contact form for you...

──────────────────── Tool: write_file ────────────────────
╭─ ✓ Result ────────────────────────────────────────────────╮
│ File written successfully: ContactForm.cshtml              │
╰────────────────────────────────────────────────────────────╯

See Getting Started Guide for detailed walkthrough with examples.

Verified Features

All features have been tested with dotnet run -- test:

Commands

CLI Commands

andy chat                       # Start interactive chat
andy chat --no-mcp              # Chat without MCP server
andy chat -c ./src              # Start with specific context path
andy run "prompt"               # Run single prompt and exit
andy config list                # View all configuration
andy config providers           # View AI provider status
andy config set key value       # Set configuration value
andy config get key             # Get configuration value
andy config validate            # Validate all API keys
andy config validate <provider> # Validate specific provider
andy config add-provider <name> <key> --baseurl <url> --model <model>
                                # Add custom OpenAI-compatible provider
andy config remove <provider>   # Remove a provider's API key
andy test                       # Run validation tests

Slash Commands (in chat)

Available Tools

AndyAI has 6 built-in tools that work offline:

When connected to the Noundry MCP server, additional platform-specific tools become available.

How It Works

AndyAI implements a Claude Code-like agentic loop:

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│ User Input  │────▶│ LLM Model   │────▶│ Response    │
└─────────────┘     └──────┬──────┘     └─────────────┘
                          │
                          ▼ (if tool calls detected)
                   ┌──────────────┐
                   │ Extract Tool │
                   │ Calls        │
                   └──────┬───────┘
                          │
                          ▼
                   ┌──────────────┐
                   │ Execute      │
                   │ Tools        │
                   └──────┬───────┘
                          │
                          ▼
                   ┌──────────────┐
                   │ Feed Results │──────┐
                   │ Back to LLM  │      │
                   └──────────────┘      │
                          ▲              │
                          └──────────────┘
                          (loop until done)

The model outputs structured tool calls:

<tool_call>
{"name": "read_file", "arguments": {"path": "Program.cs"}}
</tool_call>

AndyAI extracts these, executes the tools, and feeds results back.

Configuration

AI Providers (BYOK - Bring Your Own Key)

AndyAI supports multiple AI providers. Configure using CLI commands or appsettings.json.

Built-in Providers

Configure via CLI

# Set API key
andy config set anthropic.apikey sk-ant-api03-xxx
andy config set openai.apikey sk-xxx

# Set active provider
andy config set provider anthropic

# View all providers
andy config providers

# Validate API keys
andy config validate              # Validate all
andy config validate anthropic    # Validate specific

# Remove invalid/expired key
andy config remove anthropic

Configure via Environment Variables

export ANTHROPIC_API_KEY=sk-ant-xxx
export OPENAI_API_KEY=sk-xxx
export NEBIUS_API_KEY=your-key

Adding Custom OpenAI-Compatible Providers

Any API that follows the OpenAI /chat/completions format can be added via CLI:

# Add Nebius
andy config add-provider nebius $NEBIUS_API_KEY \
    --baseurl https://api.tokenfactory.nebius.com/v1 \
    --model openai/gpt-oss-120b

# Add Together AI
andy config add-provider together $TOGETHER_KEY \
    -b https://api.together.xyz/v1 \
    -m meta-llama/Llama-3-70b-chat-hf

# Add local Ollama
andy config add-provider ollama not-needed \
    -b http://localhost:11434/v1 \
    -m llama3

# Then activate it
andy config set provider nebius

See Provider Configuration Guide for detailed setup.

API Key Validation

Validate your API keys to ensure they're working:

# Validate all configured providers
andy config validate

# Output:
# ╭───────────┬───────────┬──────────────────────────┬───────────────────────────╮
# │ Provider  │ Status    │ Model                    │ Message                   │
# ├───────────┼───────────┼──────────────────────────┼───────────────────────────┤
# │ anthropic │ ✓ Valid   │ claude-opus-4-5-20251101 │ API key is valid          │
# │ openai    │ ✗ Invalid │ gpt-4o                   │ Invalid API key           │
# ╰───────────┴───────────┴──────────────────────────┴───────────────────────────╯

If a key is invalid or expired, remove it:

andy config remove openai

General Configuration

Config file: appsettings.json

{
  "ModelProvider": "Hosted",
  "LocalModel": {
    "Path": "models/codellama-7b-instruct.Q4_K_M.gguf",
    "ContextSize": 4096,
    "Temperature": 0.7,
    "GpuLayers": -1
  },
  "Mcp": {
    "ServerUrl": "https://mcp.noundry.ai/mcp",
    "Enabled": true
  }
}

View all settings:

andy config list

GPU Acceleration

AndyAI automatically uses GPU acceleration when:

1. NVIDIA GPU is installed with 6GB+ VRAM
2. CUDA Toolkit 12.x is installed (Download)
3. GpuLayers is set to -1 in appsettings.json (default)

Verify GPU is active:

Model initialized successfully - Backend: CUDA, GPU Layers: -1

Configure GPU layers:

{
  "LocalModel": {
    "GpuLayers": -1  // -1 = all layers on GPU, 0 = CPU only
  }
}

See System Requirements for detailed GPU setup.

Requirements

Minimum

• .NET 9.0 SDK
• 8GB+ RAM
• ~8GB disk space (model + application)

For GPU Acceleration (Recommended)

• NVIDIA GPU with 6GB+ VRAM
• CUDA Toolkit 12.x (Download)
• NVIDIA Driver 525+

See System Requirements for detailed setup instructions.

Project Structure

AndyAI/
├── Commands/           # CLI commands (chat, run, config, test)
├── Services/           # Core services (LLM, MCP, file, process)
├── UI/                 # Terminal UI (Spectre.Console)
├── Tests/              # Validation tests
├── scripts/            # Model download scripts
└── docs/               # Documentation

Validation

Run all tests to verify your installation:

dotnet run -- test

Expected output:

=== Tool Call Parsing Test Results ===
[PASS] SingleToolCall
[PASS] MultipleToolCalls
[PASS] ComplexArguments
[PASS] SpecialCharacters
[PASS] NoToolCalls
[PASS] MalformedToolCall
[PASS] NewlinesInJson
7/7 tests passed

=== Validation Test Results ===
[PASS] FileSystemService
[PASS] ProcessService
[PASS] ToolExecutor
[PASS] McpService
[PASS] AgenticConversationService
[PASS] ToolCallParsing
6/6 tests passed

Documentation

• Getting Started Guide - Detailed walkthrough with examples
• Provider Configuration - AI provider setup and custom endpoints
• System Requirements - Hardware, GPU, and CUDA setup

License

MIT License