Translate and collect data from AI applications instrumented with third-party libraries

Enable a translator to convert your telemetry data to the OpenTelemetry format and send the data to Splunk Observability Cloud.

If you've already instrumented your AI application with a third-party instrumentation library, you can enable a translator to convert the telemetry data to the OpenTelemetry format and send the data to Splunk Observability Cloud.

This solution sends traces and metrics from instrumented AI applications to Splunk Observability Cloud without requiring changes to upstream code.

LangSmith translator

Prerequisites

To translate and collect data from AI applications instrumented with LangSmith, you must meet the following requirements.
  • You have installed a LangSmith instrumentation SDK to instrument your AI application.
  • You have installed the Splunk Distribution for OpenTelemetry Generative AI packages using the following command:
    CODE 
    pip install  splunk-otel-util-genai opentelemetry-sdk>=1.31.1 langsmith >= 0.6.2 langchain>=1.2.3 langchain-core >= 1.2.7
langchain_openai >= 1.1.7 python-dotenv openai>=2.6.0
Steps
  1. Install the LangSmith translator package:
    CODE 
    pip install splunk-otel-util-genai-translator-langsmithRu

  2. Run or modify the example application from GitHub to automatically translate LangSmith SDK spans to GenAI semantic conventions. When you run this example:

    1. The LangSmith SDK creates spans with LangSmith-specific attributes (such as gen_ai.prompt, gen_ai.completion, langsmith.metadata.ls_provider, langsmith.metadata.ls_temperature, etc.).

    2. The LangsmithSpanProcessor automatically intercepts the LangSmith-specific keys and mutates it with the corresponding GenAI-compliant attributes.

    3. The translation happens transparently and does not require modifying any application code. The processor runs the translation automatically without any additional code changes.
    4. The original LangSmith-specific keys remain in the span.

Key mapping reference: LangSmith and GenAI attributes

The LangSmith translator scans invocation attributes for keys beginning with gen_ai. and langsmith. and adds the following corresponding GenAI keys:
LangSmith key GenAI mapped key
gen_ai.prompt gen_ai.input.messages
gen_ai.content.prompt gen_ai.input.messages
gen_ai.completion gen_ai.output.messages
gen_ai.content.completion gen_ai.output.messages
gen_ai.output_messages gen_ai.output.messages
langsmith.entity.input gen_ai.input.messages
langsmith.entity.output gen_ai.output.messages
langsmith.metadata.ls_provider gen_ai.system
langsmith.metadata.ls_model_name gen_ai.request.model
langsmith.metadata.ls_model_type gen_ai.operation.name
langsmith.model_name gen_ai.request.model
langsmith.provider gen_ai.system
langsmith.run_type gen_ai.operation.name
langsmith.metadata.ls_temperature gen_ai.request.temperature
langsmith.metadata.ls_max_tokens gen_ai.request.max_tokens
langsmith.metadata.ls_top_p gen_ai.request.top_p
langsmith.metadata.ls_top_k gen_ai.request.top_k
langsmith.metadata.ls_presence_penalty gen_ai.request.presence_penalty
langsmith.metadata.ls_frequency_penalty gen_ai.request.frequency_penalty
langsmith.metadata.ls_seed gen_ai.request.seed
langsmith.metadata.ls_stop_sequences gen_ai.request.stop_sequences
langsmith.temperature gen_ai.request.temperature
langsmith.max_tokens gen_ai.request.max_tokens
langsmith.top_p gen_ai.request.top_p
langsmith.top_k gen_ai.request.top_k
gen_ai.token.usage.input gen_ai.usage.input_tokens
gen_ai.token.usage.output gen_ai.usage.output_tokens
gen_ai.token.usage.total gen_ai.usage.total_tokens
langsmith.token_usage.prompt_tokens gen_ai.usage.input_tokens
langsmith.token_usage.completion_tokens gen_ai.usage.output_tokens
langsmith.token_usage.total_tokens gen_ai.usage.total_tokens
langsmith.usage.prompt_tokens gen_ai.usage.input_tokens
langsmith.usage.completion_tokens gen_ai.usage.output_tokens
langsmith.usage.total_tokens gen_ai.usage.total_tokens
langsmith.session_id gen_ai.conversation.id
langsmith.parent_run_id gen_ai.parent_run.id
langsmith.agent.name gen_ai.agent.name
langsmith.agent.type gen_ai.agent.type
langsmith.agent.description gen_ai.agent.description
langsmith.workflow.name gen_ai.workflow.name
langsmith.chain.name gen_ai.workflow.name
langsmith.error gen_ai.error.message
langsmith.error.type gen_ai.error.type
langsmith.status gen_ai.response.status
langsmith.response.model gen_ai.response.model
langsmith.response.id gen_ai.response.id
langsmith.finish_reason gen_ai.response.finish_reasons

OpenLit translator

Prerequisites

To translate and collect data from AI applications instrumented with OpenLit, you must meet the following requirements.

  • You have installed an OpenLit instrumentation SDK to instrument your AI application.

  • You have installed the Splunk Distribution for OpenTelemetry Generative AI packages using the following command:
    CODE 
    pip install  splunk-otel-util-genai opentelemetry-sdk>=1.31.1 openlit>=1.35.0 python-dotenv openai>=2.6.0

Steps

  1. Install the OpenLit translator package:
    CODE 
    pip install splunk-otel-util-genai-translator-openlit

  2. Run or modify the example application from GitHub to automatically translate OpenLit SDK spans to GenAI semantic conventions. When you run this example:

    1. The OpenLit SDK creates spans with OpenLit-specific attributes (such as gen_ai.prompt, gen_ai.completion, gen_ai.llm.provider, etc.).

    2. The OpenLitSpanProcessor automatically intercepts the OpenLit-specific spans and mutates it with the corresponding GenAI-compliant attributes.

    3. The translation happens transparently and does not require modifying any application code. After you initialize OpenLit, the processor runs the translation automatically without any additional code changes.

Key mapping reference: OpenLit and GenAI attributes

The OpenLit translator scans invocation attributes for keys beginning with gen_ai. and adds the following corresponding GenAI keys:
OpenLit key GenAI mapped key
gen_ai.completion.0.content gen_ai.output.messages
gen_ai.prompt.0.content gen_ai.input.messages
gen_ai.prompt gen_ai.input.messages
gen_ai.completion gen_ai.output.messages
gen_ai.content.prompt gen_ai.input.messages
gen_ai.content.completion gen_ai.output.messages
gen_ai.request.embedding_dimension gen_ai.embeddings.dimension.count
gen_ai.token.usage.input gen_ai.usage.input_tokens
gen_ai.token.usage.output gen_ai.usage.output_tokens
gen_ai.llm.provider gen_ai.provider.name
gen_ai.llm.model gen_ai.request.model
gen_ai.llm.temperature gen_ai.request.temperature
gen_ai.llm.max_tokens gen_ai.request.max_tokens
gen_ai.llm.top_p gen_ai.request.top_p
gen_ai.operation.type gen_ai.operation.name
gen_ai.output_messages gen_ai.output.messages
gen_ai.session.id gen_ai.conversation.id
gen_ai.tool.args gen_ai.tool.call.arguments
gen_ai.tool.result gen_ai.tool.call.result
gen_ai.vectordb.name gen_ai.vectordb.name
gen_ai.vectordb.search.query db.query.text
gen_ai.vectordb.search.results_count db.response.returned_rows

Traceloop translator

The Traceloop translator promotes legacy traceloop.* attributes attached to an LLMInvocation into OpenTelemetry GenAI semantic convention (or forward-looking custom gen_ai.*) attributes before the standard semantic convention span emitter runs.

Prerequisites

To translate and collect data from AI applications instrumented with Traceloop, you must meet the following requirements.

  • If you installed the Splunk Distribution of the OpenTelemetry Python instrumentation for LangChain or LangGraph, uninstall the instrumentation:
    CODE 
    pip uninstall splunk-otel-instrumentation-langchain

  • You have installed the OpenAI SDK.

  • You have installed a Traceloop instrumentation SDK to instrument your AI application.

  • You have installed the Splunk Distribution for OpenTelemetry Generative AI packages using the following command:
    CODE 
    pip install  splunk-otel-util-genai opentelemetry-sdk>=1.31.1 traceloop-sdk>=0.47.4 python-dotenv openai>=2.6.0

Steps

  1. Install the Traceloop translator package:
    CODE 
    pip install splunk-otel-util-genai-translator-traceloop

  2. Run or modify the example application from GitHub to automatically translate Traceloop SDK spans to GenAI semantic conventions. When you run this example:

    1. The Traceloop SDK creates spans with Traceloop-specific attributes (such as traceloop.workflow.name, traceloop.entity.name, etc.).

    2. The TraceloopSpanProcessor automatically intercepts the Traceloop-specific spans and mutates it with the corresponding GenAI-compliant attributes.

    3. The translation happens transparently and does not require modifying any application code. After you initialize Traceloop, the processor runs the translation automatically without any additional code changes.

Key mapping reference: Traceloop and GenAI attributes

The Traceloop translator scans invocation.attributes for keys beginning with traceloop. and adds the following corresponding GenAI keys:

Traceloop key GenAI mapped key Description
traceloop.entity.name gen_ai.agent.name Approximates entity as agent name.
traceloop.entity.input gen_ai.input.messages Serialized form.
traceloop.entity.output gen_ai.output.messages Serialized form.
traceloop.correlation.id gen_ai.conversation.id Approximate mapping.

Custom key mapping reference: workflow and entity hierarchy

To align Traceloop telemetry data with the OpenTelemetry entity data model, the Traceloop translator scans invocation.attributes for keys beginning with traceloop. and adds the following corresponding custom GenAI keys:

Traceloop key GenAI mapped key
traceloop.workflow.name gen_ai.workflow.name
traceloop.association.properties gen_ai.association.properties
traceloop.entity.path gen_ai.workflow.path
traceloop.span.kind gen_ai.span.kind
traceloop.entity.version gen_ai.workflow.version

Next steps

To finish setting up AI Agent Monitoring, proceed to the next step in Set up AI Agent Monitoring.