volra-observe 0.1.0

Framework-agnostic LLM observability for Volra agents

volra-observe

Framework-agnostic LLM observability for Volra agents.

Install

pip install volra-observe

Quick Start

import volra_observe

# Auto-patches OpenAI and Anthropic SDKs, starts Prometheus server on :9101
volra_observe.init(port=9101)

Metrics Exposed

Metric	Type	Labels
volra_llm_tokens_total	Counter	model, type (input/output)
volra_llm_cost_dollars_total	Counter	model
volra_llm_request_duration_seconds	Histogram	model, status
volra_llm_errors_total	Counter	model, error_type
volra_tool_calls_total	Counter	tool_name

Manual Instrumentation

from volra_observe import track_llm, llm_context

# Decorator
@track_llm(model="gpt-4o")
def call_llm(prompt):
    return client.chat.completions.create(model="gpt-4o", messages=[...])

# Context manager
with llm_context(model="claude-sonnet-4-20250514") as ctx:
    response = anthropic.messages.create(...)
    ctx["response"] = response