TypeSafe AI Tracing (Python)

Looking for TypeScript? See the TypeScript guide.

This module provides OpenInference instrumentation for the TypeSafe AI Python SDK (typesafe-sdk). Calls to TypeSafeClient.system_one and AsyncTypeSafeClient.system_one are captured as OpenInference LLM spans. A System One request sends a state plus a map of typed questions (Noul, Choice, Score) and returns one typed answer per question, so the span records the request state/model/questions as input.value, the response answers/usage as output.value, the request and resolved model names, and prompt/completion/total token counts. Requires typesafe-sdk >= 0.6.0.

Install

pip install openinference-instrumentation-typesafe typesafe-sdk arize-phoenix-otel

Setup

Use the register function to connect your application to Phoenix:

from phoenix.otel import register

tracer_provider = register(
  project_name="typesafe-app",
  auto_instrument=True,
)

Run TypeSafe AI

A simple TypeSafe AI application that is now instrumented:

from typesafe_sdk import Choice, Noul, Score, TypeSafeClient

client = TypeSafeClient()
response = client.system_one(
    state={"document": "I was charged twice. Please fix this ASAP."},
    questions={
        "billing": Noul(instructions="Is this ticket about billing?"),
        "tone": Choice(
            instructions="What is the customer's tone?",
            criteria={"calm": None, "frustrated": None, "angry": None},
        ),
        "urgency": Score(
            instructions="How urgent is this ticket?",
            criteria=["can wait", "this week", "today"],
        ),
    },
)
print(response.nouls["billing"].noul)
print(response.choices["tone"].choice)
print(response.scores["urgency"].score)

The AsyncTypeSafeClient is instrumented the same way, and questions can be passed as SDK objects (as above) or as raw dictionaries.

Observe

With instrumentation enabled, each system_one call shows up in Phoenix as an LLM span containing:

input.value: the request body (state, model, questions) as JSON
llm.invocation_parameters: the call configuration, meaning the model and any extra_body fields
output.value: the response body (model, answers, usage) as JSON
llm.request.model_name and llm.response.model_name (the resolved model)
llm.token_count.prompt, llm.token_count.completion, and llm.token_count.total

A System One call is not a chat exchange, so the state and the answers are recorded only as input.value and output.value, not as llm.input_messages / llm.output_messages.

Configuration

Because the state and the questions are recorded only in input.value, TraceConfig(hide_inputs=True) keeps the whole request off the span, and hide_outputs=True does the same for the answers. llm.invocation_parameters holds no request content, only the model and any extra_body fields; mask it with hide_llm_invocation_parameters if those are sensitive.

from openinference.instrumentation import TraceConfig
from openinference.instrumentation.typesafe import TypeSafeAIInstrumentor
from phoenix.otel import register

tracer_provider = register(project_name="typesafe-app")

TypeSafeAIInstrumentor().instrument(
    tracer_provider=tracer_provider,
    config=TraceConfig(hide_inputs=True, hide_outputs=True),
)

Tracing can also be suppressed for a block of code with suppress_tracing(), and context attributes such as using_session, using_user, and using_attributes propagate session, user, metadata, and tag information onto the spans it produces.

Developer Tools

MCP

LLM Providers

TypeScript

Python

Java

Platform

Sandboxes

Evaluation Integrations

Coding Agents

TypeSafe AI Tracing (Python)

Install

Setup

Run TypeSafe AI

Observe

Configuration

Resources

​Install

​Setup

​Run TypeSafe AI

​Observe

​Configuration

​Resources

Install

Setup

Run TypeSafe AI

Observe

Configuration

Resources