kalibr 1.1.2a0

Unified LLM Observability & Multi-Model AI Integration Framework - Deploy to GPT, Claude, Gemini, Copilot with full telemetry

Kalibr Python SDK

Production-grade observability for LLM applications. Automatically instrument OpenAI, Anthropic, and Google AI SDKs with zero code changes.

Features

• Zero-code instrumentation - Automatic tracing for OpenAI, Anthropic, and Google AI
• Cost tracking - Real-time cost calculation for all LLM calls
• Token monitoring - Track input/output tokens across providers
• Parent-child traces - Automatic trace relationship management
• Multi-provider support - Works with GPT-4, Claude, Gemini, and more

Installation

pip install kalibr

Quick Start

Auto-instrumentation (Recommended)

Simply import kalibr at the start of your application - all LLM calls are automatically traced:

import kalibr  # Enable auto-instrumentation
import openai

# Set your Kalibr API key
import os
os.environ["KALIBR_API_KEY"] = "your-kalibr-api-key"

# All OpenAI calls are now automatically traced
client = openai.OpenAI()
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}]
)

Manual Tracing with Decorator

For more control, use the @trace decorator:

from kalibr import trace
import openai

@trace(operation="summarize", provider="openai", model="gpt-4o")
def summarize_text(text: str) -> str:
    client = openai.OpenAI()
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": "Summarize the following text."},
            {"role": "user", "content": text}
        ]
    )
    return response.choices[0].message.content

result = summarize_text("Your long text here...")

Multi-Provider Example

import kalibr
import openai
import anthropic

# OpenAI call - automatically traced
openai_client = openai.OpenAI()
gpt_response = openai_client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Explain quantum computing"}]
)

# Anthropic call - automatically traced
anthropic_client = anthropic.Anthropic()
claude_response = anthropic_client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Explain machine learning"}]
)

Configuration

Configure the SDK using environment variables:

Variable	Description	Default
KALIBR_API_KEY	API key for authentication	Required
KALIBR_COLLECTOR_URL	Collector endpoint URL	http://localhost:8001/api/ingest
KALIBR_TENANT_ID	Tenant identifier for multi-tenant setups	default
KALIBR_WORKFLOW_ID	Workflow identifier for grouping traces	default
KALIBR_SERVICE_NAME	Service name for OpenTelemetry spans	kalibr-app
KALIBR_ENVIRONMENT	Environment (prod, staging, dev)	prod
KALIBR_AUTO_INSTRUMENT	Enable/disable auto-instrumentation	true
KALIBR_CONSOLE_EXPORT	Enable console span export for debugging	false

CLI Tools

The SDK includes command-line tools for running and deploying applications:

# Run your app locally with tracing
kalibr serve myapp.py

# Run with managed runtime lifecycle
kalibr run myapp.py --port 8000

# Deploy to cloud platforms
kalibr deploy myapp.py --runtime fly.io

# Fetch trace data by ID
kalibr capsule <trace-id>

Supported Providers

Provider	Models	Auto-Instrumentation
OpenAI	GPT-4, GPT-4o, GPT-3.5	Yes
Anthropic	Claude 3 Opus, Sonnet, Haiku	Yes
Google	Gemini Pro, Gemini Flash	Yes

Examples

See the examples/ directory for complete examples:

• basic_example.py - Simple tracing example
• basic_agent.py - Agent with auto-instrumentation
• advanced_example.py - Advanced tracing patterns
• cross_vendor.py - Multi-provider workflows
• test_mas.py - Multi-agent system demonstration

Development

# Clone the repository
git clone https://github.com/kalibr-systems/kalibr-sdk-python.git
cd kalibr-sdk-python

# Install in development mode
pip install -e ".[dev]"

# Run tests
pytest

# Format code
black kalibr/
ruff check kalibr/

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.

Links

• Documentation
• GitHub Issues
• PyPI Package