Anthropic Provider
The Anthropic provider implements support for Claude models through the Anthropic API, including Claude 3 Sonnet, Opus, and Haiku models.
Supported Models
python
from aicore.models_metadata import METADATA
# List available Anthropic models
anthropic_models = [model for model in METADATA if model.startswith("anthropic-")]
print(anthropic_models)Key Features
- Automatic token counting and cost tracking
- Extended thinking mode for complex reasoning
- Prompt caching control
- Async-first implementation
Configuration
python
from aicore.llm.config import LlmConfig
config = LlmConfig(
provider="anthropic",
api_key="your_api_key",
model="claude-3-7-sonnet",
temperature=0.7,
max_tokens=4000,
thinking=True # Enable extended reasoning
)Basic Usage
python
from aicore.llm import Llm
# Initialize LLM with Anthropic config
llm = Llm(config=config)
# Simple completion
response = llm.complete("Explain quantum computing in simple terms")
print(response)Advanced Usage
Thinking Mode
python
# Enable thinking mode for step-by-step reasoning
thinking_config = LlmConfig(
provider="anthropic",
api_key="your_api_key",
model="claude-3-7-sonnet",
thinking=True
)
llm = Llm(config=thinking_config)
response = llm.complete("Solve this math problem: (3x + 5 = 20)")Model Specifications
| Model Name | Context Window | Best For |
|---|---|---|
| claude-3-7-sonnet | 200K tokens | Complex reasoning tasks |
| claude-3-5-haiku | 200K tokens | Fast, cost-effective |
Additional Notes
- Requires Anthropic API key
- Supports both synchronous and asynchronous operations
- Implements automatic retries for rate limits
- Detailed usage metrics available through observability system
For advanced configuration options, see the base provider documentation.