Skip to main content
SDK

PUVINoise SDK for AI agents

Independent OpenTelemetry SDK with tracing and observability for Anthropic Claude, OpenAI, Ollama, and other LLM providers — plus optional decision telemetry for behaviour intelligence.

Overview

The puvinoise-sdk package instruments agent runs and exports OTLP traces to your PUVINoise collector. It is designed for multi-tenant production: every span carries tenant and agent identity so the control plane can filter and scope correctly. Optional decision APIs emit intent, tool selection, reasoning checkpoints, and state transitions for the behaviour pipeline.

Reference document

Install and quickstart

From the SDK README. Configure via environment variables, then wrap agent work in a traced run.

Install
pip install puvinoise-sdk — optional extras [anthropic], [openai], or [all].
Bootstrap
bootstrap() reads PUVINOISE_* and CUST_LLM_PROVIDER from the environment.
Trace
run_with_trace creates an agent.run span and exports on the batch schedule.
Migrate
Schema v1 and v2 are both accepted; upgrades never reject telemetry by version.
# Install the PUVINoise SDK
pip install puvinoise-sdk
 
# Optional provider extras
pip install puvinoise-sdk[anthropic]
pip install puvinoise-sdk[openai]
pip install puvinoise-sdk[all]
Related resources

Documentation

Full environment variable contract and resource attribute map.

Documentation

Guides

Instrument, verify in Command Centre, then migrate schemas safely.

Guides

Developers hub

Surfaces overview: CLI, API, OpenTelemetry, sidecar, frameworks.

Developers

Ready to instrument your agents?

Start with the quickstart above, or talk to us about your framework and tenancy model.