moduna 0.1.6

Python OpenTelemetry helpers for Moduna AI traces.

moduna

Python OpenTelemetry helpers for Moduna AI traces.

moduna makes Moduna tracing easy to add with one import and one instantiation. It configures OpenTelemetry once, exports traces to Moduna over OTLP HTTP, and provides ergonomic helpers for Python LangChain applications.

Installation

Install from PyPI:

pip install moduna

For LangChain callback support, install the optional LangChain dependency:

pip install "moduna[langchain]"

The PyPI distribution is moduna; the Python import package is moduna_otel.

When working from this repository, use uv:

uv sync --extra dev --extra langchain

Getting Started

Set your Moduna API key with an environment variable:

export MODUNA_API_KEY="md_live_..."

Create the SDK once near application startup:

from moduna_otel import ModunaOTEL

otel = ModunaOTEL(
    agent_name="support-agent",
    framework="langchain",
)

You can also pass the API key directly:

from moduna_otel import ModunaOTEL

otel = ModunaOTEL(
    agent_name="support-agent",
    framework="langchain",
    api_key="md_live_...",
)

Multiple ModunaOTEL(...) instances share the same OpenTelemetry provider, so creating per-module wrappers will not create duplicate exporters.

LangChain

Create a LangChain callback handler and pass it into your model or chain call:

from moduna_otel import ModunaOTEL

otel = ModunaOTEL(
    agent_name="support-agent",
    framework="langchain",
)

handler = otel.langchain_handler(
    {
        "conversation_id": "conversation-123",
        "session_id": "session-456",
    }
)

result = chain.invoke(
    {"input": "Summarize this ticket"},
    config={
        "callbacks": [handler],
        "metadata": {
            "conversationId": "conversation-123",
            "sessionId": "session-456",
        },
    },
)

The handler records LLM/chat model spans, prompts, completions, token usage when available, streamed token events, Moduna conversation/session attributes, and errors.

For applications that use global LangChain callbacks, registration is available as a best-effort helper:

otel.register_global_langchain_handler(
    {
        "conversation_id": "conversation-123",
        "session_id": "session-456",
    }
)

LangChain global callback APIs vary across versions. If global registration is not available in your installed LangChain version, Moduna logs one warning and continues without breaking application code.

Manual Instrumentation

Use instrument() to wrap custom model calls or application work in a Moduna CLIENT span:

from moduna_otel import ModunaOTEL

otel = ModunaOTEL(agent_name="support-agent", framework="langchain")


def call_model(span):
    span.set_attribute("app.operation", "ticket-summary")
    return model.invoke("Summarize this ticket")


result = otel.instrument(
    "support-agent.summary",
    call_model,
    {
        "conversation_id": "conversation-123",
        "session_id": "session-456",
    },
)

If the callback raises an exception, the exception is recorded on the span and then re-raised.

Configuration

ModunaOTEL accepts keyword arguments:

ModunaOTEL(
    agent_name="support-agent",
    framework="langchain",  # Python currently supports LangChain
    api_key=None,           # defaults to MODUNA_API_KEY
    headers=None,           # extra OTLP HTTP headers
    auto_shutdown=True,     # flush on process exit
)

It also accepts a mapping:

otel = ModunaOTEL(
    {
        "agentName": "support-agent",
        "framework": "langchain",
        "autoShutdown": True,
    }
)

Development

Install development dependencies:

uv sync --extra dev --extra langchain

Run tests:

uv run pytest

Format code:

uv run ruff format

Run the linter:

uv run ruff check

Build source and wheel distributions:

uv build

The built artifacts are written to dist/.

Contributing

Contributions are welcome. Please keep changes focused and aligned with the TypeScript SDK behavior.

Before opening a pull request:

1. Add or update focused pytest coverage for behavior changes.
2. Run uv run pytest.
3. Run uv run ruff format and uv run ruff check.
4. Run uv build for packaging changes.
5. Keep telemetry failures non-fatal to user application code.
6. Avoid broad refactors unless they are required for the change.

Useful areas for contribution include LangChain version compatibility, additional token-usage extraction cases, and documentation examples for common framework integrations.