AI Provider Setup Guide

Choose Your AI Provider - OpenAI, Anthropic, or Local Models

The dal ai commands support multiple AI providers. Choose what works best for you:

Option 1: OpenAI (GPT-4, GPT-3.5)

Best for: Production use, highest quality responses

Setup

Get API Key
- Go to https://platform.openai.com/api-keys
- Click "Create new secret key"
- Copy the key (starts with sk-proj- or sk-)

Set Environment Variable

export OPENAI_API_KEY="sk-proj-..."

# Add to your shell config for persistence
echo 'export OPENAI_API_KEY="sk-proj-..."' >> ~/.zshrc
source ~/.zshrc

Use DAL

dal ai code "Create a DeFi lending protocol"
dal ai explain mycontract.dal
dal ai audit token.dal

Configuration

# Choose model (default: gpt-4)
export OPENAI_MODEL="gpt-4"          # Most capable
export OPENAI_MODEL="gpt-3.5-turbo"  # Faster, cheaper

Pricing

GPT-4: ~0.03/1Kinputtokens, 0.06/1K output tokens
GPT-3.5: ~0.0015/1Kinputtokens, 0.002/1K output tokens

Option 2: Anthropic (Claude)

Best for: Long context, detailed analysis, code review

Setup

Get API Key
- Go to https://console.anthropic.com/
- Navigate to "API Keys"
- Create new key
- Copy the key (starts with sk-ant-)

Set Environment Variable

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

# Add to your shell config for persistence
echo 'export ANTHROPIC_API_KEY="sk-ant-..."' >> ~/.zshrc
source ~/.zshrc

Use DAL

dal ai code "Create a token contract"
dal ai review complex_system.dal
dal ai audit defi_protocol.dal

Configuration

# Choose model (default: claude-3-5-sonnet-20241022)
export ANTHROPIC_MODEL="claude-3-5-sonnet-20241022"  # Most capable
export ANTHROPIC_MODEL="claude-3-opus-20240229"     # Highest intelligence
export ANTHROPIC_MODEL="claude-3-haiku-20240307"    # Fastest, cheapest

Pricing

Claude 3.5 Sonnet: ~0.003/1Kinputtokens, 0.015/1K output tokens
Claude 3 Opus: ~0.015/1Kinputtokens, 0.075/1K output tokens
Claude 3 Haiku: ~0.00025/1Kinputtokens, 0.00125/1K output tokens

Option 3: Local Models (Ollama, LM Studio)

Best for: Privacy, offline use, free unlimited usage

Setup with Ollama

Install Ollama

# macOS
brew install ollama

# Linux
curl -fsSL https://ollama.com/install.sh | sh

# Windows
# Download from https://ollama.com/download

Start Ollama Server
```
ollama serve
```

Download a Model

# Code generation models
ollama pull codellama         # 7B, good for code
ollama pull deepseek-coder    # Excellent for coding
ollama pull phind-codellama   # Optimized for code

# General purpose models
ollama pull llama2            # Good all-around
ollama pull mistral           # Fast and capable
ollama pull llama3            # Latest, most capable

# Small/fast models
ollama pull tinyllama         # Very fast

Configure DAL

export DAL_AI_ENDPOINT="http://localhost:11434/api/generate"
export DAL_AI_MODEL="codellama"  # or deepseek-coder, llama2, etc.

# Add to shell config
echo 'export DAL_AI_ENDPOINT="http://localhost:11434/api/generate"' >> ~/.zshrc
echo 'export DAL_AI_MODEL="codellama"' >> ~/.zshrc
source ~/.zshrc

Use DAL (Offline!)

dal ai code "Create a REST API"
dal ai explain myfile.dal
dal ai test contract.dal

Model Recommendations

Model	Size	Speed	Quality	Best For
`codellama`	7B	Fast	Good	Code generation
`deepseek-coder`	6.7B	Fast	Excellent	Code, best quality
`llama3`	8B	Medium	Excellent	General purpose
`mistral`	7B	Fast	Good	General purpose
`phind-codellama`	34B	Slow	Excellent	Complex code (needs GPU)

Pricing

FREE! Runs locally on your machine
No API costs
No rate limits
Works offline

Using Multiple Providers

You can configure all three and DAL will choose automatically:

# Set all three
export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-ant-..."
export DAL_AI_ENDPOINT="http://localhost:11434/api/generate"

# Priority order: OpenAI > Anthropic > Local > Fallback
dal ai code "test"  # Uses OpenAI (first priority)

Override Priority

# Use Anthropic instead of OpenAI
OPENAI_API_KEY="" dal ai code "test"

# Use local model
OPENAI_API_KEY="" ANTHROPIC_API_KEY="" dal ai code "test"

# Force fallback mode (no AI)
OPENAI_API_KEY="" ANTHROPIC_API_KEY="" DAL_AI_ENDPOINT="" dal ai code "test"

Switching Providers

Team A: OpenAI Only

# .env file
OPENAI_API_KEY=sk-proj-...

Team B: Anthropic Only

# .env file
ANTHROPIC_API_KEY=sk-ant-...

Team C: Local Only (Privacy-focused)

# .env file
DAL_AI_ENDPOINT=http://localhost:11434/api/generate
DAL_AI_MODEL=codellama

Individual Developer: Mix and Match

# Use OpenAI for code generation (fast)
export OPENAI_API_KEY="sk-..."

# But use local for security audits (private)
dal ai code "Create API"           # Uses OpenAI
OPENAI_API_KEY="" dal ai audit contract.dal  # Uses local (private)

Testing Your Setup

Test OpenAI

export OPENAI_API_KEY="your-key"
dal ai code "print hello world"

Test Anthropic

export ANTHROPIC_API_KEY="your-key"
dal ai code "print hello world"

Test Local

ollama serve
export DAL_AI_ENDPOINT="http://localhost:11434/api/generate"
export DAL_AI_MODEL="codellama"
dal ai code "print hello world"

Troubleshooting

"No API key found"

Check environment variables: echo $OPENAI_API_KEY
Make sure to export the variable
Try restarting your terminal

"OpenAI API error 401"

Invalid API key
Check key format: should start with sk-proj- or sk-
Regenerate key at https://platform.openai.com/api-keys

"Connection refused" (Local)

Make sure Ollama is running: ollama serve
Check endpoint: curl http://localhost:11434/api/generate
Try different port if 11434 is blocked

"Rate limit exceeded"

OpenAI: Upgrade your account or wait
Anthropic: Check your usage limits
Local: No rate limits!

Cost Comparison

Small Project (100 requests/day)

Provider	Model	Monthly Cost
OpenAI	GPT-3.5	~$5
OpenAI	GPT-4	~$50
Anthropic	Claude Haiku	~$2
Anthropic	Claude Sonnet	~$15
Anthropic	Claude Opus	~$75
Local	Any	$0

Medium Project (1000 requests/day)

Provider	Model	Monthly Cost
OpenAI	GPT-3.5	~$50
OpenAI	GPT-4	~$500
Anthropic	Claude Haiku	~$20
Anthropic	Claude Sonnet	~$150
Local	Any	$0

Recommendation

Development/Testing: Use local models (free)
Production/High Quality: Use OpenAI GPT-4 or Claude Sonnet
Budget Conscious: Use Claude Haiku or GPT-3.5
Privacy Sensitive: Use local models only

Quick Reference

# OpenAI
export OPENAI_API_KEY="sk-proj-..."
export OPENAI_MODEL="gpt-4"  # optional

# Anthropic
export ANTHROPIC_API_KEY="sk-ant-..."
export ANTHROPIC_MODEL="claude-3-5-sonnet-20241022"  # optional

# Local (Ollama)
export DAL_AI_ENDPOINT="http://localhost:11434/api/generate"
export DAL_AI_MODEL="codellama"  # optional

# Test
dal ai code "hello world"

Security Best Practices

Never commit API keys to git

echo ".env" >> .gitignore
echo "*.key" >> .gitignore

Use environment variables, not hard-coded keys

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

# Bad - NEVER do this
# let api_key = "sk-..."

Rotate keys regularly
- Regenerate API keys every 90 days
- Delete old keys after rotation

Use separate keys for dev/prod

# Development
export OPENAI_API_KEY="sk-dev-..."

# Production
export OPENAI_API_KEY="sk-prod-..."

Monitor usage
- Check OpenAI usage: https://platform.openai.com/usage
- Check Anthropic usage: https://console.anthropic.com/

Support

Need Help?

Full integration guide: docs/development/AI_API_INTEGRATION.md
CLI reference: docs/CLI_QUICK_REFERENCE.md
Phase 3 documentation: docs/development/CLI_PHASE3_COMPLETE.md

Can't get API keys?

Use local models (Ollama) - completely free and offline
All commands work in fallback mode (basic templates)

Want to contribute?

Add support for more providers (Cohere, HuggingFace, etc.)
Improve prompts for better code generation
Add caching to reduce costs