LLM API Configuration Guide

April 6, 2026 · View on GitHub

The deep analysis system supports both Anthropic Claude and OpenAI APIs. Choose based on your availability, cost, and preference.

🎯 Quick Setup

Option 1: Anthropic Claude (Default)

# Set your Anthropic API key
export ANTHROPIC_API_KEY="sk-ant-..."

Default Model: claude-sonnet-4-20250514

Optional: Specify a different model

export ANTHROPIC_MODEL="claude-opus-4-20250514"

Option 2: OpenAI

# Set provider to OpenAI
export LLM_API="openai"

# Set your OpenAI API key
export OPENAI_API_KEY="sk-..."

Default Model: gpt-4o

Optional: Specify a different model

export OPENAI_MODEL="gpt-4o-mini"

📋 Supported Models

Anthropic Claude

  • claude-sonnet-4-20250514 (default, good balance)
  • claude-opus-4-20250514 (highest quality)
  • claude-haiku-4-20250514 (fastest, lowest cost)

OpenAI

  • gpt-4o (default, good balance)
  • gpt-4o-mini (faster, lower cost)
  • gpt-4-turbo (legacy, still good)
  • gpt-3.5-turbo (fastest, lowest quality)

🔄 Switching Between APIs

# Use Anthropic
export LLM_API="anthropic"  # or just don't set (it's the default)
export ANTHROPIC_API_KEY="sk-ant-..."

# Use OpenAI
export LLM_API="openai"
export OPENAI_API_KEY="sk-..."

Method 2: Per-Command

# Run with Anthropic
LLM_API=anthropic ANTHROPIC_API_KEY="sk-ant-..." node tools/deep-analyzer.mjs "Geoffrey Hinton"

# Run with OpenAI
LLM_API=openai OPENAI_API_KEY="sk-..." node tools/deep-analyzer.mjs "Geoffrey Hinton"

🔧 Advanced Configuration

Custom Base URL (for OpenAI-compatible APIs)

# Using Azure OpenAI
export LLM_API="openai"
export OPENAI_BASE_URL="https://your-resource.openai.azure.com/"
export OPENAI_API_KEY="your-key"
export OPENAI_MODEL="gpt-4"

# Using other OpenAI-compatible APIs
export LLM_API="openai"
export OPENAI_BASE_URL="https://api.deepseek.com/v1"
export OPENAI_API_KEY="your-key"
export OPENAI_MODEL="deepseek-chat"

💰 Cost Comparison

Per Mentor Analysis (3 papers + 3 public sources)

ProviderModelInput TokensOutput TokensEst. Cost
Anthropicclaude-sonnet-4~9,000~3,000~$0.06
Anthropicclaude-haiku-4~9,000~3,000~$0.006
OpenAIgpt-4o~9,000~3,000~$0.05
OpenAIgpt-4o-mini~9,000~3,000~$0.003
OpenAIgpt-3.5-turbo~9,000~3,000~$0.002

Estimates as of 2026-04-01. Actual costs may vary.

📝 Example Usage

Analyzing a Mentor with Different APIs

# Using Anthropic Claude
export ANTHROPIC_API_KEY="sk-ant-..."
node tools/deep-analyzer.mjs "Geoffrey Hinton" --affiliation "University of Toronto"

# Switch to OpenAI
export LLM_API="openai"
export OPENAI_API_KEY="sk-..."
node tools/deep-analyzer.mjs "Geoffrey Hinton" --affiliation "University of Toronto"

Generating Skills with Different APIs

# Using Anthropic (default)
export ANTHROPIC_API_KEY="sk-ant-..."
node tools/skill-generator.mjs "Geoffrey Hinton" --affiliation "University of Toronto" --deep-analyze

# Using OpenAI
export LLM_API="openai"
export OPENAI_API_KEY="sk-..."
node tools/skill-generator.mjs "Geoffrey Hinton" --affiliation "University of Toronto" --deep-analyze

⚙️ Permanent Configuration

For Bash/Zsh Users

Add to ~/.bashrc or ~/.zshrc:

# Anthropic Claude (default)
export ANTHROPIC_API_KEY="sk-ant-..."
export ANTHROPIC_MODEL="claude-sonnet-4-20250514"

# OR OpenAI
export LLM_API="openai"
export OPENAI_API_KEY="sk-..."
export OPENAI_MODEL="gpt-4o"

Then reload:

source ~/.bashrc  # or source ~/.zshrc

For Project-Specific Configuration

Create .env file in project root:

# .env
LLM_API=openai
OPENAI_API_KEY=sk-...
OPENAI_MODEL=gpt-4o

Then load before running:

source .env
node tools/deep-analyzer.mjs "Geoffrey Hinton"

🧪 Testing Configuration

Test which API is being used:

# Should print: anthropic or openai
node -e "import('./tools/content-analyzer.mjs').then(m => console.log(m.getAPIConfig().provider))"

Test a simple analysis:

# Create test file
echo "Test paper content" > /tmp/test_paper.txt

# Run with Anthropic
export ANTHROPIC_API_KEY="sk-ant-..."
node tools/content-analyzer.mjs /tmp/test_paper.txt "Test Paper" "Test Mentor"

# Run with OpenAI
export LLM_API=openai OPENAI_API_KEY="sk-..."
node tools/content-analyzer.mjs /tmp/test_paper.txt "Test Paper" "Test Mentor"

🚨 Troubleshooting

Error: ANTHROPIC_API_KEY not set

Solution: Set the API key

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

Error: OPENAI_API_KEY not set

Solution: Set the API key and provider

export LLM_API="openai"
export OPENAI_API_KEY="sk-..."

Error: API error 401

Cause: Invalid API key Solution: Verify your API key is correct and active

Error: API error 429

Cause: Rate limit exceeded Solution: Wait a few minutes and try again, or switch to a different API provider

Error: Could not parse as JSON

Cause: Model didn't return valid JSON Solution: Try a different model (e.g., gpt-4o instead of gpt-3.5-turbo)

🎓 Recommendation

For Best Results: Anthropic Claude

  • Better at following complex instructions
  • More structured output
  • Better at analysis tasks
  • Use: claude-sonnet-4-20250514 or claude-opus-4-20250514

For Cost Efficiency: OpenAI

  • Lower cost per token
  • Faster response times
  • Good for simple tasks
  • Use: gpt-4o or gpt-4o-mini

📚 Additional Resources

🔐 Security Best Practices

  1. Never commit API keys to git

    # Add to .gitignore
echo ".env" >> .gitignore
echo "*.key" >> .gitignore

  2. Use environment variables, not hardcoded keys

  3. Rotate keys regularly (monthly recommended)

  4. Set spending limits in your API console

  5. Monitor usage to detect unusual activity

Last updated: 2026-04-01