Integrations

Spanlight reads traces that your application sends it. This section shows how to send them from the libraries you already use, so you add tracing without rewriting your code. Each card shows the language the guide is written for.

Built in

Two ways to send traces need no framework guide. They work with any stack, and the guides below build on them.

Python SDKLanguage: Python. Wrap the OpenAI and Anthropic clients, trace your own functions and set sessions and users.

OpenTelemetry (OTLP)Language: Any language. Send OpenTelemetry GenAI spans from any exporter over OTLP/HTTP.

Providers

Send the calls you make to a model provider. Each guide covers the Python SDK, the gateway and OpenTelemetry.

OpenAILanguage: Python, JavaScript, cURL. Trace Chat Completions calls with the Python wrapper, or send them through the gateway from any language.

AnthropicLanguage: Python, JavaScript, cURL. Trace Claude Messages calls with the Python wrapper, or send them through the gateway from any language.

OpenTelemetry exporters

Send the spans an OpenTelemetry instrumentation library already writes. Each guide shows the exporter setup its library’s recording was made with, and links to OTLP compatibility, which lists field by field what each library captures and loses.

Vercel AI SDKLanguage: TypeScript. Export generateText, streamText and tool calls from a Node.js or Next.js server, on AI SDK 5, 6 and 7.

LiteLLMLanguage: Python. Trace litellm.completion calls with LiteLLM's own OpenTelemetry callback, in either of its modes.

OpenLLMetry (Traceloop)Language: Python. Trace workflows, tasks, agents, tools and OpenAI calls with the Traceloop SDK.

OpenInferenceLanguage: Python. Trace the OpenAI client and LangChain with the instrumentations behind Arize Phoenix.

OpenAI Agents SDKLanguage: Python. Trace agent runs, turns, tool calls and model calls with the OpenInference instrumentation.

LangGraphLanguage: Python. Trace graph nodes, tools and model calls with the OpenInference LangChain instrumentation.

Pydantic AILanguage: Python. Trace agent runs, tool calls and model calls with Pydantic AI's built-in instrumentation, no Logfire needed.

LlamaIndexLanguage: Python. Trace agent workflow steps, tools and model calls with the OpenInference LlamaIndex instrumentation.

CI

Check a prompt change before it merges.

GitHub ActionLanguage: GitHub Actions workflow (YAML). Run a prompt over a dataset on every pull request and fail the job when a score drops below a threshold.

How the guides are tested

Every integration guide opens with a “Verified” note. It gives the date the instructions were last run and the exact package versions they were run with, and the install commands on the page use the same versions. If you use a different version, the steps may differ.

Each guide also lists what Spanlight records from that library and what it does not, so you know what the dashboard can show before you rely on it.

Related pages

Ways to send dataCompare the Python SDK, OTLP and the gateway.

OTLP attribute mappingWhich OpenTelemetry attribute fills which Spanlight field.

OTLP compatibilityWhat each recorded exporter library and version captures and loses.

TroubleshootingFind out why no trace arrived.