Core Concepts

Before diving into implementation, it helps to understand the foundational concepts that power Brokle. These concepts form the building blocks for monitoring, evaluating, and improving your AI applications.

The Observability Stack

Brokle provides a complete observability stack for AI applications:

Key Concepts

How They Connect

Understanding how these concepts relate helps you design effective observability:

Hierarchy

1. Sessions group multiple traces from the same user or conversation
2. Traces capture complete request-response cycles
3. Spans represent individual operations within a trace
4. Evaluations attach quality metrics to spans or traces
5. Cost Analytics aggregate spending across all levels

Core Principles

1. Minimal Code Changes

Brokle integrates with your existing code without major refactoring:

from brokle import Brokle, wrap_openai
import openai

client = Brokle(api_key="bk_...")
openai_client = wrap_openai(openai.OpenAI(), brokle=client)

# Your existing code works unchanged
response = openai_client.chat.completions.create(...)

2. OpenTelemetry Compatible

Brokle is built on OpenTelemetry standards, ensuring:

• Interoperability with existing observability tools
• Standardized trace and span formats
• Easy integration with your infrastructure

3. Privacy First

• Sensitive data masking options
• On-premise deployment available
• No data leaves your infrastructure (self-hosted)

4. Real-Time Insights

• Live trace streaming
• Instant cost calculations
• Real-time quality scoring

Quick Start Path

Based on your needs, here's where to start:

Next Steps

• Create your first trace →
• View the quickstart guide →
• Explore integrations →