Skip to main content

​
Configuration

LLM Magic can be configured through environment variables and the config/llm-magic.php configuration file.

​
Environment Variables

Add these variables to your .env file to configure LLM Magic:

​
API Keys

Configure API keys for the LLM providers you want to use: 
# OpenAI (default provider)
OPENAI_API_KEY=sk-...
OPENAI_ORGANIZATION_ID=org-...  # Optional

# Anthropic (Claude)
ANTHROPIC_API_KEY=sk-ant-...

# Google Gemini
GEMINI_API_KEY=...

# Mistral AI
MISTRAL_API_KEY=...

# OpenRouter (access to multiple models)
OPENROUTER_API_KEY=...

# Together AI
TOGETHERAI_API_KEY=...

# DeepSeek
DEEPSEEK_API_KEY=...

# Groq
GROQ_API_KEY=...
You only need to configure API keys for the providers you plan to use. LLM Magic will throw an error if you try to use a provider without its API key configured.

​
Model Configuration

Set default models for different use cases: 
# Default model for all operations
LLM_MAGIC_MODEL=openai/gpt-4o-mini

# Model for cost-sensitive operations
LLM_MAGIC_CHEAP_MODEL=openai/gpt-4o-mini

# Model specifically for data extraction
LLM_MAGIC_EXTRACTION_MODEL=anthropic/claude-3-5-sonnet-20241022

# Model for chat operations
LLM_MAGIC_CHAT_MODEL=google/gemini-2.0-flash-lite

# Model for generating embeddings
LLM_MAGIC_EMBEDDINGS_MODEL=openai/text-embedding-3-small

​
Extraction Settings

Configure parallel extraction behavior: 
# Number of concurrent extraction operations (default: 3)
LLM_MAGIC_EXTRACTION_CONCURRENCY=3

# Default maximum tokens per model (default: 10000)
LLM_MAGIC_DEFAULT_MAX_TOKENS=10000

​
Artifact Storage

Configure where processed artifacts (PDFs, images, etc.) are stored: 
# Base path for artifacts (default: storage/app/magic-artifacts)
LLM_MAGIC_ARTIFACTS_BASE=/path/to/artifacts

# Laravel disk for artifacts (default: artifacts)
LLM_MAGIC_ARTIFACTS_DISK=artifacts

# Prefix for artifact paths (default: empty)
LLM_MAGIC_ARTIFACTS_PREFIX=llm-magic/

​
Python Integration

LLM Magic uses Python for some operations (like PDF processing): 
# Python working directory
LLM_MAGIC_PYTHON_CWD=/path/to/python

# Use UV for Python package management (default: true)
LLM_MAGIC_PYTHON_USE_UV=true

# Path to UV executable (default: /usr/bin/env uv)
LLM_MAGIC_PYTHON_UV_PATH=/usr/local/bin/uv

# Path to Python binary (default: auto-detected)
LLM_MAGIC_PYTHON_BIN_PATH=/path/to/python

​
Configuration File

The config/llm-magic.php file provides full control over LLM Magic’s behavior. Here’s the complete configuration structure: 
<?php

return [
    'llm' => [
        'default' => env('LLM_MAGIC_MODEL', 'openai/gpt-4o-mini'),
    ],
    'models' => [
        'default' => env('LLM_MAGIC_MODEL', 'openai/gpt-4o-mini'),
        'cheap' => env('LLM_MAGIC_CHEAP_MODEL', 'openai/gpt-4o-mini'),
        'extraction' => env('LLM_MAGIC_EXTRACTION_MODEL', null),
        'chat' => env('LLM_MAGIC_CHAT_MODEL', null),
        'embeddings' => env('LLM_MAGIC_EMBEDDINGS_MODEL', 'openai/text-embedding-3-small'),
    ],
];

​
Model Selection

LLM Magic supports multiple model providers and formats:

​
Available Providers

  • OpenAI: openai/gpt-4o, openai/gpt-4o-mini, openai/gpt-4-turbo
  • Anthropic: anthropic/claude-3-5-sonnet-20241022, anthropic/claude-3-opus
  • Google: google/gemini-2.0-flash-lite, google/gemini-pro
  • Mistral: mistral/mistral-large, mistral/mistral-small
  • OpenRouter: Access to 100+ models via openrouter/provider/model-name
  • Together AI: togetherai/meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo
  • DeepSeek: deepseek/deepseek-chat

​
Accessing Available Models

Get a list of all available models programmatically: 
use Mateffy\Magic;

// Get all registered models
$models = Magic::models();

// Get the default model name
$defaultName = Magic::defaultModelName();

// Get the default model label
$defaultLabel = Magic::defaultModelLabel();

// Get the default model instance
$model = Magic::defaultModel();

​
Custom Extraction Strategies

Register custom extraction strategies for specialized use cases: 
use Mateffy\Magic;
use Mateffy\Magic\Extraction\Strategies\Strategy;

Magic::registerStrategy('my-custom-strategy', MyCustomStrategy::class);
Built-in strategies include:
  • simple: Single-pass extraction
  • sequential: Process artifacts sequentially
  • sequential-auto-merge: Sequential with automatic merging
  • parallel: Process artifacts in parallel
  • parallel-auto-merge: Parallel with automatic merging
  • double-pass: Two-pass extraction for better accuracy
  • double-pass-auto-merge: Double-pass with automatic merging

​
Next Steps

Chat API Reference

Learn about all chat configuration options

Extraction Strategies

Choose the right extraction strategy for your use case

Model Providers

Explore all supported LLM providers

Advanced Usage

Learn advanced patterns and techniques

Build docs developers (and LLMs) love

Get started for freeTalk to us