Python SDK for AI agent monitoring, LLM cost tracking, benchmarking, and more. Integrates with most LLMs and agent frameworks including CrewAI, Agno, OpenAI Agents SDK, Langchain, Autogen, AG2, and CamelAI
Python agentagentopsagents-sdkaianthropicautogencost-estimationcrewaievalsevaluation-metricsgroqlangchainllmmistralollamaopenaiopenai-agents
import pytest
from pathlib import Path
import os
from vcr.record_mode import RecordMode
@pytest.fixture(scope="session")
def vcr_config():
"""Configure VCR.py for recording HTTP interactions.
This fixture sets up VCR.py with:
- YAML serialization
- Cassette storage in fixtures/recordings
- Comprehensive header filtering for API keys and sensitive data
- Request matching on URI, method, and body
"""
# Define cassette storage location
vcr_cassettes = Path(__file__).parent / "recordings"
vcr_cassettes.mkdir(parents=True, exist_ok=True)
# Define sensitive headers to filter
sensitive_headers = [
# Generic API authentication
("authorization", "REDACTED"),
("x-api-key", "REDACTED"),
("api-key", "REDACTED"),
("bearer", "REDACTED"),
# AgentOps API keys
("x-agentops-api-key", "REDACTED"),
# LLM service API keys
("openai-api-key", "REDACTED"),
("anthropic-api-key", "REDACTED"),
("cohere-api-key", "REDACTED"),
("x-cohere-api-key", "REDACTED"),
("ai21-api-key", "REDACTED"),
("x-ai21-api-key", "REDACTED"),
("replicate-api-token", "REDACTED"),
("huggingface-api-key", "REDACTED"),
("x-huggingface-api-key", "REDACTED"),
("claude-api-key", "REDACTED"),
("x-claude-api-key", "REDACTED"),
("x-railway-request-id", "REDACTED"),
("X-Railway-Request-Id", "REDACTED"),
# Authentication tokens
("x-api-token", "REDACTED"),
("api-token", "REDACTED"),
("x-auth-token", "REDACTED"),
("x-session-token", "REDACTED"),
# OpenAI specific headers
("openai-organization", "REDACTED"),
("x-request-id", "REDACTED"),
("__cf_bm", "REDACTED"),
("_cfuvid", "REDACTED"),
("cf-ray", "REDACTED"),
# Rate limit headers
("x-ratelimit-limit-requests", "REDACTED"),
("x-ratelimit-limit-tokens", "REDACTED"),
("x-ratelimit-remaining-requests", "REDACTED"),
("x-ratelimit-remaining-tokens", "REDACTED"),
("x-ratelimit-reset-requests", "REDACTED"),
("x-ratelimit-reset-tokens", "REDACTED"),
# Mistral headers
("x-mistral-api-key", "REDACTED"),
# Groq headers
("x-groq-api-key", "REDACTED"),
# LiteLLM headers
("x-litellm-api-key", "REDACTED"),
# Ollama headers
("x-ollama-api-key", "REDACTED"),
# TaskWeaver headers
("x-taskweaver-api-key", "REDACTED"),
# Additional provider version headers
("anthropic-version", "REDACTED"),
("cohere-version", "REDACTED"),
("x-stainless-lang", "REDACTED"),
("x-stainless-arch", "REDACTED"),
("x-stainless-os", "REDACTED"),
("x-stainless-async", "REDACTED"),
("x-stainless-runtime", "REDACTED"),
("x-stainless-runtime-version", "REDACTED"),
]
def filter_response_headers(response):
"""Filter sensitive headers from response."""
headers = response["headers"]
headers_lower = {k.lower(): k for k in headers} # Map of lowercase -> original header names
for header, replacement in sensitive_headers:
header_lower = header.lower()
if header_lower in headers_lower:
# Replace using the original header name from the response
original_header = headers_lower[header_lower]
headers[original_header] = replacement
return response
vcr_config = {
"filter_headers": [
"authorization",
"Authorization",
"X-OpenAI-Client-User-Agent",
],
"match_on": ["method", "scheme", "host", "port", "path", "query"],
"record_mode": RecordMode.ONCE if os.getenv("CI") else RecordMode.NEW_EPISODES,
"path_transformer": lambda path: path.replace("\\", "/"),
"record_on_exception": False,
}
return vcr_config