observio 0.1.0

Observio Python SDK

Observio Python SDK

Observio is an open-source, full-stack LLM observability and evaluation platform. It helps you trace agentic workflows, capture nested tool invocations, log input/output tokens, and attribute costs across multiple AI models seamlessly.

Features

• Full-Stack Tracing: Capture backend, tool, and LLM layers with zero friction.
• Dynamic Auto-Decorators: Decorate functions using @observe() to automatically start and stop tracing spans.
• OpenTelemetry Standard: Built on standard OpenTelemetry semantic conventions for AI.
• Cost & Token Attribution: Track precise model pricing and token counts on spans.

Installation

Install the SDK via pip:

pip install observio

Quick Start

1. Initialization

Initialize the SDK at the start of your application:

from observio import Observio

Observio.initialize(
    project_api_key="your-observio-project-api-key",
    base_url="http://localhost:8000",        # Base URL of your local or hosted Observio instance
    force_http=True                          # Set to True for local development
)

2. Tracing Functions

Use the @observe() decorator to trace any function or RAG component:

from observio import observe

@observe()
def call_llm(prompt: str):
    # Traced under a single span
    return "This response is traced"

@observe()
def process_rag_pipeline(query: str):
    context = retrieve_context(query)
    response = call_llm(query)
    return response

3. Adding Span Attributes

Enhance your spans with rich context metadata:

from opentelemetry import trace
from observio import observe

@observe()
def embed_text(text: str):
    span = trace.get_current_span()
    # Categorize the layer
    span.set_attribute("observio.layer", "ai")
    # Log usage metrics
    span.set_attribute("gen_ai.usage.input_tokens", len(text) // 4)
    span.set_attribute("lmnr.span.total_cost", 0.0001)
    return [0.1, 0.2, 0.3]