OpenLLMetry is an Apache-2.0 set of OpenTelemetry-based extensions, built and maintained by Traceloop, that gives complete observability over Python GenAI and LLM applications, and it is aimed at teams already running LLM workloads that need tracing, metrics, and model monitoring without adopting a proprietary agent.

What it is

OpenLLMetry is a set of extensions built on top of OpenTelemetry, and it provides complete observability over an LLM application. The repository contains standard OpenTelemetry instrumentations for LLM providers and Vector DBs, as well as a Traceloop SDK that makes it easy to get started while still outputting standard OpenTelemetry data that can be connected to an existing observability stack. The project is built and maintained by Traceloop under the Apache 2.0 license, is written in Python, and lives in the AI & Machine Learning / Machine Learning Infrastructure category. A separate JS/TS implementation, OpenLLMetry-JS, exists for JavaScript and TypeScript applications.

The concrete problem it solves is that LLM calls, vector database lookups, and model interactions are opaque to conventional application tracing, so teams end up with hand-rolled logging around provider clients instead of a uniform signal. OpenLLMetry supplies the instrumentation layer for those calls and emits it as OpenTelemetry data, which means traces and metrics land in observability tooling that a team already operates. Because it uses OpenTelemetry under the hood, it can be connected to existing observability solutions such as Datadog and Honeycomb rather than forcing a new backend.

Key capabilities

• Standard OpenTelemetry instrumentations for LLM providers and Vector DBs, so model and retrieval calls are traced without custom spans.
• The traceloop-sdk package with Traceloop.init(), which instruments an application in a single call.
• Traceloop.init(disable_batch=True) for local runs, so traces appear immediately instead of waiting on batch sending.
• Standard OpenTelemetry output that can be pointed at many destinations, including OpenTelemetry Collector, Datadog, Honeycomb, Grafana, Dynatrace, New Relic, Splunk, SigNoz, Sentry, Axiom, Azure Application Insights, Google Cloud, Oracle Cloud, Tencent Cloud, IBM Instana, KloudMate, Laminar, Braintrust, Dash0, Highlight, HyperDX, Scorecard, Service Now Cloud Observability, and Traceloop itself.
• Direct use of individual instrumentations for applications that already have OpenTelemetry configured, without adopting the full SDK.
• Semantic conventions for LLM observability that are now part of OpenTelemetry, produced through the upstream community process.
• A parallel JS/TS implementation, OpenLLMetry-JS, for non-Python services.

Who uses it and how

• Python teams instrumenting an LLM application who want traces of provider calls and vector database queries flowing into an existing observability stack.
• Organizations already running OpenTelemetry that add OpenLLMetry instrumentations directly instead of a second SDK.
• Teams whose traces must land in an incumbent vendor such as Datadog, Honeycomb, New Relic, Dynatrace, Grafana, or Splunk, using the supported destination for each.
• Developers working locally who set disable_batch=True to see traces as they are produced.
• Contributors: the topics include good-first-issue, good-first-issues, and help-wanted, and the repository carries 7,437 stars and 1,083 forks.

Getting started

Install the SDK with pip install traceloop-sdk, then add from traceloop.sdk import Traceloop and Traceloop.init() to the application. Full setup guidance lives in the docs at traceloop.com/docs/openllmetry/getting-started-python.

How it compares

OpenLLMetry sits on the OpenTelemetry standard rather than replacing the backend, so it competes with the instrumentation layer of proprietary LLM monitoring agents while leaving the storage and query layer to the vendor a team already uses. Its semantic conventions have been contributed upstream to OpenTelemetry, which keeps the trace schema portable across destinations, and the sibling OpenLLMetry-JS project extends the same approach to JavaScript and TypeScript services.

When to use it — and when not to

A self-hoster must operate an OpenTelemetry pipeline and whichever destination backend is chosen, since OpenLLMetry produces telemetry rather than storing or displaying it. Teams that want a single hosted product with no collector, exporter configuration, or backend to run should not pick it. The repository currently shows 676 open issues, last push on 2026-09-16, and the readme excerpt is truncated with substantive setup detail deferred to an external documentation site, so assess triage load and documentation depth before committing.

pip install traceloop-sdk

from traceloop.sdk import Traceloop

Traceloop.init()

Traceloop.init(disable_batch=True)