Multi-Provider Support: Choose Your AI Backend

CCJK is designed to be provider-agnostic. While optimized for Claude, it supports multiple AI providers, giving you flexibility in choosing the best model for your needs.

Supported Providers

Provider	Models	Best For
Anthropic	Claude 3.5 Sonnet, Claude 3 Opus	Complex reasoning, code generation
OpenAI	GPT-4, GPT-4 Turbo, GPT-3.5	General tasks, wide compatibility
Google	Gemini Pro, Gemini Ultra	Multi-modal tasks
Local	Ollama, LM Studio	Privacy, offline use
Azure	Azure OpenAI	Enterprise compliance

Configuration

Basic Provider Setup

hljs yaml
# .claude/config.yaml
providers:
  default: anthropic

  anthropic:
    api_key: ${ANTHROPIC_API_KEY}
    model: claude-sonnet-4-20250514
    max_tokens: 8192

  openai:
    api_key: ${OPENAI_API_KEY}
    model: gpt-4-turbo-preview
    max_tokens: 4096

  google:
    api_key: ${GOOGLE_API_KEY}
    model: gemini-pro

  local:
    endpoint: http://localhost:11434
    model: codellama:13b

Environment Variables

hljs bash
# .env or shell profile
export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export GOOGLE_API_KEY="..."

Switching Providers

Command Line

hljs bash
# Use default provider
ccjk

# Specify provider
ccjk --provider openai

# Specify model
ccjk --provider anthropic --model claude-3-opus-20240229

In-Session Switching

You: /provider openai
Switched to OpenAI (gpt-4-turbo-preview)

You: /provider anthropic
Switched to Anthropic (claude-sonnet-4-20250514)

You: /model claude-3-opus-20240229
Switched to claude-3-opus-20240229

Per-Task Provider

hljs yaml
# .claude/skills/complex-analysis.yaml
name: complex-analysis
provider: anthropic
model: claude-3-opus-20240229  # Use Opus for complex tasks

prompt: |
  Perform deep analysis of...

Provider-Specific Features

Anthropic (Claude)

Best for:

• Complex code generation
• Multi-file refactoring
• Nuanced code review

Configuration:

hljs yaml
anthropic:
  model: claude-sonnet-4-20250514
  max_tokens: 8192
  features:
    extended_thinking: true  # For complex problems
    artifacts: true          # For structured output

OpenAI (GPT-4)

Best for:

• Quick tasks
• Wide language support
• Function calling

Configuration:

hljs yaml
openai:
  model: gpt-4-turbo-preview
  max_tokens: 4096
  features:
    json_mode: true
    function_calling: true
    vision: true  # For GPT-4V

Google (Gemini)

Best for:

• Multi-modal tasks
• Long context windows
• Google Cloud integration

Configuration:

hljs yaml
google:
  model: gemini-pro
  max_tokens: 8192
  features:
    multi_modal: true
    long_context: true

Local Models (Ollama)

Best for:

• Privacy-sensitive code
• Offline development
• Cost savings

Configuration:

hljs yaml
local:
  endpoint: http://localhost:11434
  model: codellama:13b
  options:
    num_ctx: 4096
    temperature: 0.7

Fallback Configuration

Automatic Fallback

hljs yaml
providers:
  default: anthropic

  fallback:
    - provider: openai
      condition: rate_limit
    - provider: local
      condition: api_error

  anthropic:
    model: claude-sonnet-4-20250514
    rate_limit_fallback: openai

  openai:
    model: gpt-4-turbo-preview

Cost-Based Routing

hljs yaml
routing:
  # Use cheaper models for simple tasks
  simple_tasks:
    provider: openai
    model: gpt-3.5-turbo

  # Use powerful models for complex tasks
  complex_tasks:
    provider: anthropic
    model: claude-3-opus-20240229

  # Use local for sensitive code
  sensitive:
    provider: local
    model: codellama:13b

Setting Up Local Models

Ollama Setup

hljs bash
# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull a coding model
ollama pull codellama:13b

# Or a general model
ollama pull llama2:13b

# Start the server
ollama serve

LM Studio Setup

1. Download LM Studio from lmstudio.ai
2. Download a model (e.g., CodeLlama, Mistral)
3. Start the local server
4. Configure CCJK:

hljs yaml
local:
  endpoint: http://localhost:1234/v1
  model: local-model
  api_type: openai_compatible

Enterprise Configuration

Azure OpenAI

hljs yaml
azure:
  endpoint: https://your-resource.openai.azure.com
  api_key: ${AZURE_OPENAI_KEY}
  api_version: "2024-02-15-preview"
  deployment: your-gpt4-deployment

AWS Bedrock

hljs yaml
bedrock:
  region: us-east-1
  model: anthropic.claude-3-sonnet-20240229-v1:0
  credentials:
    access_key: ${AWS_ACCESS_KEY}
    secret_key: ${AWS_SECRET_KEY}

Private Deployment

hljs yaml
private:
  endpoint: https://ai.internal.company.com
  api_key: ${INTERNAL_API_KEY}
  model: company-model-v2
  tls:
    ca_cert: /path/to/ca.crt
    client_cert: /path/to/client.crt

Comparing Providers

Performance Comparison

Run benchmarks:

hljs bash
ccjk benchmark --providers anthropic,openai,local --task code-review

Output:

Provider Benchmark Results
==========================

Task: Code Review (500 lines)

| Provider  | Model              | Time   | Quality | Cost    |
|-----------|--------------------| -------|---------|---------|
| Anthropic | claude-3.5-sonnet  | 12.3s  | 9.2/10  | $0.045  |
| OpenAI    | gpt-4-turbo        | 15.1s  | 8.8/10  | $0.062  |
| Local     | codellama:13b      | 28.4s  | 7.1/10  | $0.00   |

Cost Analysis

hljs bash
ccjk cost --period month --by-provider

Monthly Cost Analysis
=====================

Anthropic:  $45.20 (1,200 requests)
OpenAI:     $12.50 (450 requests)
Local:      $0.00  (800 requests)

Total:      $57.70
Estimated savings from local: $28.00

Best Practices

1. Match Model to Task

hljs yaml
task_routing:
  # Quick questions → fast, cheap model
  quick:
    provider: openai
    model: gpt-3.5-turbo

  # Code generation → balanced model
  code:
    provider: anthropic
    model: claude-sonnet-4-20250514

  # Architecture decisions → powerful model
  architecture:
    provider: anthropic
    model: claude-3-opus-20240229

  # Sensitive code → local model
  sensitive:
    provider: local
    model: codellama:13b

2. Set Spending Limits

hljs yaml
limits:
  daily_spend: 10.00
  monthly_spend: 200.00
  alert_threshold: 0.8  # Alert at 80%

  per_request:
    max_tokens: 4096
    max_cost: 0.50

3. Monitor Usage

hljs bash
# View usage statistics
ccjk stats --period week

# Export for analysis
ccjk stats --export csv --output usage.csv

4. Test Before Switching

hljs bash
# Test a provider before making it default
ccjk test-provider openai --task "Review this code..."

# Compare outputs
ccjk compare --providers anthropic,openai --task "Implement..."

Troubleshooting

Provider Connection Issues

hljs bash
# Test connectivity
ccjk diagnose --provider anthropic

# Check API key
ccjk verify-key --provider openai

Model Not Available

hljs yaml
# Configure fallback for unavailable models
anthropic:
  model: claude-3-opus-20240229
  fallback_model: claude-sonnet-4-20250514

Rate Limiting

hljs yaml
rate_limiting:
  retry_attempts: 3
  retry_delay: 1000  # ms
  exponential_backoff: true
  fallback_on_limit: true

Conclusion

Multi-provider support gives you flexibility to:

• Choose the best model for each task
• Manage costs effectively
• Maintain privacy with local models
• Ensure availability with fallbacks

Start with a single provider, then expand as you understand your needs.

Next: Return to Getting Started to review the basics, or explore our provider directory for real-world upstream options.