Anthropic Provider

Configure and use Anthropic Claude models with PicoClaw

Overview

The Anthropic provider enables access to Claude models from Anthropic. PicoClaw uses the native Anthropic API protocol for optimal compatibility and performance.

Configuration

Model List Format

Add Anthropic models to your model_list configuration:

{
  "model_list": [
    {
      "model_name": "claude-sonnet-4.6",
      "model": "anthropic/claude-sonnet-4.6",
      "api_key": "sk-ant-your-key",
      "api_base": "https://api.anthropic.com/v1",
      "request_timeout": 300
    }
  ],
  "agents": {
    "defaults": {
      "model_name": "claude-sonnet-4.6"
    }
  }
}

Configuration Parameters

Parameter	Type	Required	Default	Description
model_name	string	Yes	-	Alias for this model configuration
model	string	Yes	-	Model identifier with anthropic/ prefix
api_key	string	Yes*	-	Your Anthropic API key (or use OAuth)
api_base	string	No	https://api.anthropic.com/v1	API endpoint URL
request_timeout	integer	No	120	Request timeout in seconds

*Required unless using OAuth authentication

Available Models

Anthropic provides several Claude model families:

Claude 4 Series (Latest)

• anthropic/claude-opus-4 - Most capable, best for complex tasks
• anthropic/claude-sonnet-4.6 - Balanced performance and speed
• anthropic/claude-haiku-4 - Fast and efficient

Claude 3.5 Series

• anthropic/claude-opus-3.5 - Previous generation flagship
• anthropic/claude-sonnet-3.5 - Previous generation balanced model

Claude 3 Series

• anthropic/claude-opus-3 - Legacy flagship model
• anthropic/claude-sonnet-3 - Legacy balanced model
• anthropic/claude-haiku-3 - Legacy fast model

Claude models support extended context windows. Opus and Sonnet models support up to 200K tokens.

Setup Instructions

1. Get API Key

1. Visit Anthropic Console
2. Sign in or create an account
3. Navigate to API Keys section
4. Click Create Key
5. Copy your API key (starts with sk-ant-)

2. Configure PicoClaw

Edit ~/.picoclaw/config.json:

{
  "model_list": [
    {
      "model_name": "claude",
      "model": "anthropic/claude-sonnet-4.6",
      "api_key": "sk-ant-your-actual-key-here"
    }
  ],
  "agents": {
    "defaults": {
      "model_name": "claude",
      "max_tokens": 8192,
      "temperature": 0.7
    }
  }
}

3. Test Connection

picoclaw agent -m "Hello Claude, test my connection"

Advanced Configuration

Custom API Endpoint

Use a custom Anthropic-compatible endpoint:

{
  "model_name": "claude-proxy",
  "model": "anthropic/claude-sonnet-4.6",
  "api_base": "https://my-proxy.com/v1",
  "api_key": "sk-ant-...",
  "request_timeout": 300
}

Prompt Caching

Claude supports prompt caching to reduce costs for repeated prompts. PicoClaw automatically uses cache control for system messages:

{
  "model_name": "claude",
  "model": "anthropic/claude-sonnet-4.6",
  "api_key": "sk-ant-..."
}

Cache control is automatically applied to:

• System prompts
• Tool definitions
• Large context blocks

Extended Context

Claude models support large context windows:

{
  "agents": {
    "defaults": {
      "model_name": "claude",
      "max_tokens": 8192
    }
  }
}

Context limits by model:

• Opus: 200K tokens
• Sonnet: 200K tokens
• Haiku: 200K tokens

Authentication Methods

API Key (Recommended)

Standard API key authentication:

{
  "model_list": [
    {
      "model_name": "claude",
      "model": "anthropic/claude-sonnet-4.6",
      "api_key": "sk-ant-your-key"
    }
  ]
}

OAuth / Token (Advanced)

For OAuth-based authentication:

{
  "providers": {
    "anthropic": {
      "auth_method": "oauth",
      "api_base": "https://api.anthropic.com/v1"
    }
  }
}

Then authenticate:

picoclaw auth login --provider anthropic

Paste your API token when prompted.

Troubleshooting

Rate Limiting

If you encounter rate limits:

• Check your plan’s rate limits in the Anthropic Console
• Implement request throttling
• Upgrade your plan for higher limits

Invalid API Key

Error: 401 Authentication failed

• Verify your API key is correct
• Ensure key starts with sk-ant-
• Check key hasn’t been revoked in Console

Content Filtering

Some providers (like when using Claude through proxies) have content filtering:

• Rephrase your query
• Use different wording
• Contact Anthropic support if false positives occur

Timeout Errors

Increase timeout for long-running requests:

{
  "model_name": "claude",
  "model": "anthropic/claude-sonnet-4.6",
  "api_key": "sk-ant-...",
  "request_timeout": 600
}

Model Selection Guide

Use Case	Recommended Model	Notes
Complex analysis	claude-opus-4	Best reasoning capabilities
General tasks	claude-sonnet-4.6	Best balance of speed and quality
Fast responses	claude-haiku-4	Optimized for speed
Large documents	claude-sonnet-4.6	200K context window
Cost-sensitive	claude-haiku-4	Most economical option

Cost Optimization

1. Choose appropriate models: Use Haiku for simple tasks, Sonnet for general use, Opus only when needed
2. Enable prompt caching: Automatically enabled for system prompts (75% cost reduction)
3. Set max_tokens: Limit response length to reduce costs
4. Monitor usage: Check Anthropic Console regularly
5. Use extended context wisely: Only include necessary context

Legacy Configuration (Deprecated)

Older configuration format (still supported):

{
  "providers": {
    "anthropic": {
      "api_key": "sk-ant-your-key",
      "api_base": "https://api.anthropic.com/v1"
    }
  },
  "agents": {
    "defaults": {
      "provider": "anthropic",
      "model": "claude-sonnet-4.6"
    }
  }
}

Migrate to model_list format for better features like load balancing and fallbacks.

Related Documentation

• Model List Configuration
• Agent Configuration
• Prompt Caching