OpenLIT

OpenLIT instruments the model client a framework calls, so it records frameworks that emit no GenAI spans of their own. Python.

Turn tracing on

pip install openlit

Set the exporter variables from OpenTelemetry, and this one, without which OpenLIT prints spans instead of sending them:

OTEL_TRACES_EXPORTER=otlp

Then once at startup:

import openlit

openlit.init(disable_metrics=True, capture_message_content=False)

What is recorded

• Each OpenAI client call, with its model, provider and tokens, cached tokens included.
• A call that fails, as failed, with the client's error name.
• Not tool calls: OpenLIT has no tool span.
• Spans OpenLIT adds for a framework's own steps carry totals and are not counted again.

A streamed OpenAI call made without stream_options={"include_usage": True} is recorded with OpenLIT's own estimate of its tokens, which can be far from what the provider bills. Ask for usage on streamed calls.

Checked with OpenLIT 1.45 on the OpenAI client, CrewAI and LlamaIndex.

Budgets

Spans arrive after the call, so spend counts against your budgets and raises alerts, but a budget cannot stop a call recorded this way.