Authentication¶

FeatCopilot supports two LLM backends:

GitHub Copilot SDK (default) - Native integration with GitHub Copilot
LiteLLM - Universal interface supporting 100+ providers

GitHub Copilot SDK (Default)¶

The default backend uses GitHub Copilot SDK. If you have GitHub Copilot CLI installed and authenticated, it works automatically.

from featcopilot import AutoFeatureEngineer

# Uses GitHub Copilot SDK by default
engineer = AutoFeatureEngineer(
    engines=['tabular', 'llm'],
    llm_config={'model': 'gpt-5.2'}
)

LiteLLM Backend¶

For access to 100+ LLM providers, use the LiteLLM backend.

Supported Providers¶

Provider	Environment Variable	Model Example
OpenAI	`OPENAI_API_KEY`	`gpt-4o`, `gpt-4`
Azure OpenAI	`AZURE_API_KEY`, `AZURE_API_BASE`	`azure/gpt-4`
Anthropic	`ANTHROPIC_API_KEY`	`claude-3-opus`, `claude-3-sonnet`
Google	`GOOGLE_API_KEY`	`gemini-pro`
GitHub Models	`GITHUB_API_KEY`	`github/gpt-4o`, `github/Llama-3.2-11B-Vision-Instruct`
GitHub Copilot	OAuth device flow	`github_copilot/gpt-4`
AWS Bedrock	AWS credentials	`bedrock/anthropic.claude-v2`
Ollama	(local)	`ollama/llama2`

See full list of supported providers

Step 1: Install LiteLLM¶

pip install featcopilot[litellm]

Step 2: Set Up Authentication¶

Option A: Environment Variables (Recommended)¶

# OpenAI
export OPENAI_API_KEY="sk-..."

# Azure OpenAI
export AZURE_API_KEY="..."
export AZURE_API_BASE="https://your-resource.openai.azure.com/"
export AZURE_API_VERSION="2024-02-15-preview"

# Anthropic
export ANTHROPIC_API_KEY="sk-ant-..."

# GitHub Models
export GITHUB_API_KEY="ghp_..."

Option B: Pass Directly in Code¶

from featcopilot import AutoFeatureEngineer

engineer = AutoFeatureEngineer(
    engines=['tabular', 'llm'],
    llm_config={
        'model': 'gpt-4o',
        'backend': 'litellm',
        'api_key': 'sk-...',  # Not recommended for production
    }
)

Step 3: Use LiteLLM Backend¶

from featcopilot import AutoFeatureEngineer

# Specify backend='litellm' to use LiteLLM
engineer = AutoFeatureEngineer(
    engines=['llm'],
    llm_config={
        'model': 'gpt-4o',
        'backend': 'litellm'
    }
)

Provider-Specific Examples (LiteLLM)¶

OpenAI¶

from featcopilot import AutoFeatureEngineer

engineer = AutoFeatureEngineer(
    engines=['tabular', 'llm'],
    llm_config={
        'model': 'gpt-4o',
        'backend': 'litellm',
        # Uses OPENAI_API_KEY from environment
    }
)

Azure OpenAI¶

from featcopilot import AutoFeatureEngineer

engineer = AutoFeatureEngineer(
    engines=['tabular', 'llm'],
    llm_config={
        'model': 'azure/your-deployment-name',
        'backend': 'litellm',
        # Uses AZURE_API_KEY, AZURE_API_BASE from environment
    }
)

Anthropic Claude¶

from featcopilot import AutoFeatureEngineer

engineer = AutoFeatureEngineer(
    engines=['tabular', 'llm'],
    llm_config={
        'model': 'claude-3-sonnet-20240229',
        'backend': 'litellm',
        # Uses ANTHROPIC_API_KEY from environment
    }
)

GitHub Marketplace Models¶

from featcopilot import AutoFeatureEngineer

engineer = AutoFeatureEngineer(
    engines=['tabular', 'llm'],
    llm_config={
        'model': 'github/gpt-4o',  # or github/Llama-3.2-11B-Vision-Instruct
        'backend': 'litellm',
        # Uses GITHUB_API_KEY from environment
    }
)

GitHub Copilot Chat API¶

from featcopilot import AutoFeatureEngineer

# Uses OAuth device flow - you'll be prompted to authenticate
engineer = AutoFeatureEngineer(
    engines=['tabular', 'llm'],
    llm_config={
        'model': 'github_copilot/gpt-4',
        'backend': 'litellm',
    }
)

Local Models (Ollama)¶

from featcopilot import AutoFeatureEngineer

# Run Ollama locally first: ollama run llama2
engineer = AutoFeatureEngineer(
    engines=['tabular', 'llm'],
    llm_config={
        'model': 'ollama/llama2',
        'backend': 'litellm',
        'api_base': 'http://localhost:11434',
    }
)

Environment Variables¶

Common environment variables for LiteLLM:

# OpenAI
export OPENAI_API_KEY="sk-..."

# Azure OpenAI
export AZURE_API_KEY="..."
export AZURE_API_BASE="https://..."
export AZURE_API_VERSION="2024-02-15-preview"

# Anthropic
export ANTHROPIC_API_KEY="sk-ant-..."

# Custom proxy/base URL
export LITELLM_PROXY_BASE_URL="http://your-proxy:8000"

Troubleshooting¶

"API key not found"¶

Ensure your API key is set in the environment:

# Check if key is set
echo $OPENAI_API_KEY

# Set it if missing
export OPENAI_API_KEY="sk-..."

"Rate limit exceeded"¶

LLM APIs have rate limits. If exceeded:

Wait a few minutes before retrying
Reduce max_suggestions in LLM config
Use caching to avoid redundant requests

"Invalid model"¶

Check that you're using the correct model name for your provider:

# OpenAI models
'gpt-4', 'gpt-3.5-turbo', 'gpt-4-turbo'

# Azure models (use your deployment name)
'azure/my-gpt4-deployment'

# Anthropic models
'claude-3-opus-20240229', 'claude-3-sonnet-20240229'

Using Behind Corporate Proxy¶

# Set proxy environment variables
export HTTP_PROXY=http://proxy.company.com:8080
export HTTPS_PROXY=http://proxy.company.com:8080

Security Best Practices¶

Security

Never commit API keys to version control
Use environment variables for sensitive configuration
Regularly rotate API keys
Review your provider's data handling policies

Offline/Mock Mode¶

If LLM is unavailable, FeatCopilot automatically falls back to mock responses:

# This works without authentication (uses mock LLM)
engineer = AutoFeatureEngineer(engines=['tabular', 'llm'])

# You'll see this warning:
# "Warning: LLM not configured. Using mock responses."

# Features will still be generated using heuristics
X_transformed = engineer.fit_transform(X, y)

The mock mode generates context-aware features based on column names, but without the semantic understanding of a real LLM.