Python SDK

Official WatchLLM documentation for python sdk.

Python SDK#

Python SDK for AI Observability and LLM Monitoring

Complete SDK for instrumenting Python applications with WatchLLM's semantic caching and cost optimization layer. Monitor LLM calls, agent steps, errors, and performance metrics with automatic PII redaction and intelligent batching.

Quick Start#

pip install watchllm-sdk-python
from watchllm import WatchLLMClient, EventType, Status
 
client = WatchLLMClient(
    api_key='your-api-key',
    project_id='your-project-id'
)
 
# Track LLM calls
client.track({
    'type': EventType.LLM_CALL,
    'status': Status.SUCCESS,
    'model': 'gpt-4',
    'input': 'Hello, how are you?',
    'output': 'I am doing well, thank you!',
    'tokens': {'prompt': 10, 'completion': 20},
    'cost': 0.002
})

Complete Documentation#

For comprehensive guides, API reference, and advanced usage patterns, see the full Python SDK Documentation.

Key Features#

  • Thread-Safe: Designed for both synchronous and asynchronous Python applications
  • Automatic Batching: Efficient event batching for optimal performance
  • PII Redaction: Automatic sensitive data protection
  • Agent Monitoring: Track complex agent workflows and step execution
  • Error Tracking: Comprehensive error monitoring and debugging
  • Cost Estimation: Real-time cost calculation and optimization insights

Installation#

pip install watchllm-sdk-python

Or with specific package managers:

# poetry
poetry add watchllm-sdk-python
 
# conda
conda install watchllm-sdk-python

Basic Usage#

from watchllm import WatchLLMClient, Status
 
client = WatchLLMClient(
    api_key=os.getenv('WATCHLLM_API_KEY'),
    project_id=os.getenv('WATCHLLM_PROJECT_ID')
)
 
# Simple LLM call tracking
client.track({
    'type': 'llm_call',
    'status': Status.SUCCESS,
    'model': 'gpt-4',
    'input': 'What is the capital of France?',
    'output': 'The capital of France is Paris.',
    'tokens': {'prompt': 8, 'completion': 7},
    'cost': 0.0014
})

Learn More#

Visit the complete Python SDK documentation for:

  • Advanced configuration options
  • Agent workflow monitoring
  • Error handling patterns
  • Performance optimization
  • Troubleshooting guides

© 2025 WatchLLM. All rights reserved.