neatlogs 1.1.2

A Python package for extracting and managing LLM logs to build a collaborative workspace

Neatlogs

A comprehensive LLM tracking system that automatically captures and logs all LLM API calls with detailed metrics.

Features

• 🚀 Automatic LLM Call Tracking: Seamlessly tracks all LLM API calls without code changes
• 📊 Comprehensive Metrics: Token usage, costs, response times, and more
• 🔌 Multi-Provider Support: OpenAI, Anthropic, Google Gemini, Azure OpenAI, and LiteLLM
• 🔗 LangChain Integration: Seamless tracking for LangChain chains, agents, and tools
• 🧵 Session Management: Track conversations across multiple threads and agents
• 📝 Structured Logging: Detailed logs with OpenTelemetry support
• 🎯 Easy Integration: Simple one-line initialization
• 🔍 Real-time Monitoring: Live tracking and statistics

Quick Start

Installation

Basic installation (no LLM libraries):

pip install neatlogs

Basic Usage

import neatlogs

# Initialize tracking with your API key
neatlogs.init(
    api_key="your-api-key-here"
)

# add tags
neatlogs.add_tags(["neatlogs"])
# Now all LLM calls are automatically tracked!
# Use any supported LLM library normally

# Get session statistics
stats = neatlogs.get_session_stats()
print(f"Total calls: {stats['total_calls']}")
print(f"Total cost: ${stats['total_cost']:.4f}")

Supported Providers

• OpenAI (GPT models)
• Anthropic (Claude models)
• Google Gemini (Gemini models)
• Azure OpenAI
• LiteLLM (unified interface)

Framework

Neatlogs provides comprehensive support for various AI frameworks and models:

LangChain Integration

Neatlogs provides comprehensive tracking for all LangChain components and workflows:

• LLM & Chat Models: Track all LLM calls, token usage, costs, and response times
• Chains: Monitor chain execution, inputs, outputs, and performance metrics
• Agents: Capture agent actions, tool calls, decision-making processes, and reasoning
• Tools: Record tool usage, inputs, outputs, and execution times
• RAG Systems: Track retrieval-augmented generation workflows including vector searches and document retrieval
• Async Workflows: Full support for asynchronous LangChain pipelines and concurrent operations
• Error Handling: Capture and log errors across all LangChain components
• Model Detection: Automatic identification of underlying LLM models and providers.

LangChain Callback Handler

Neatlogs provides a dedicated callback handler for LangChain to enable detailed tracking of your LangChain applications without modifying your existing code.

Usage

from langchain.chains import LLMChain
from langchain.llms import OpenAI
import neatlogs_test

# Get the callback handler
handler = neatlogs_test.get_langchain_callback_handler(api_key="your-api-key")

# Use it with your LangChain components
llm = OpenAI()
chain = LLMChain(llm=llm, callbacks=[handler])

# Your chain calls will now be tracked automatically
result = chain.run("Hello world")

Features

• LLM Tracking: Captures all LLM calls with token usage, costs, and response times
• Chain Monitoring: Tracks chain executions, inputs, and outputs
• Tool Call Tracking: Monitors tool usage and performance
• Agent Monitoring: Records agent actions and decision processes
• Automatic Detection: Automatically detects model types and providers
• Async Support: Full support for both synchronous and asynchronous workflows

Asynchronous Usage

For asynchronous LangChain workflows:

from neatlogs_test.integration.callbacks.langchain.callback import AsyncNeatlogsLangchainCallbackHandler

# Use the async handler for async workflows
async_handler = AsyncNeatlogsLangchainCallbackHandler(api_key="your-api-key")

# Use with async chains
result = await async_chain.arun(..., callbacks=[async_handler])

CrewAI Integration

CrewAI is a framework for orchestrating role-playing AI agents. Neatlogs provides seamless integration with CrewAI through automatic instrumentation:

• Agent Tracking: Monitor all agent activities, tasks, and interactions
• Crew Orchestration: Track crew-level operations and agent coordination
• Task Monitoring: Capture task execution, delegation, and completion
• Automatic Setup: No code changes required - just initialize with neatlogs.init()

import neatlogs_test
from crewai import Agent, Task, Crew

# Initialize Neatlogs (that's all you need!)
neatlogs_test.init(api_key="your-api-key")

# Your CrewAI code works normally and gets tracked automatically
agent = Agent(role="Researcher", goal="Research AI trends")
task = Task(description="Research latest AI developments")
crew = Crew(agents=[agent], tasks=[task])

result = crew.kickoff()

Configuration Options

neatlogs.init(
    api_key="your-api-key",
    tags=["tag1", "tag2"],
)

Session Statistics

Get comprehensive insights into your LLM usage:

stats = neatlogs.get_session_stats()

# Available metrics:
# - total_calls: Number of LLM API calls
# - total_tokens_input: Total input tokens
# - total_tokens_output: Total output tokens
# - total_cost: Total cost in USD
# - average_response_time: Average response time
# - provider_breakdown: Usage by provider
# - model_breakdown: Usage by model