uselemma-tracing 3.0.2

OpenTelemetry-based tracing module for Lemma

uselemma-tracing

OpenTelemetry-based tracing for AI agents. Capture inputs, outputs, timing, token usage, and errors — then view everything in Lemma.

Installation

pip install uselemma-tracing

Quick Start

1. Register the tracer provider

Call register_otel once when your application starts. It reads LEMMA_API_KEY and LEMMA_PROJECT_ID from environment variables by default.

from uselemma_tracing import register_otel

register_otel()

2. Wrap your agent

agent creates a root OpenTelemetry span named ai.agent.run. Return a value from the wrapped function — the wrapper auto-captures it as ai.agent.output and closes the span automatically. The result is a TraceResult with named fields — and also supports tuple unpacking for backward compatibility.

from uselemma_tracing import TraceContext, agent

async def run_agent(user_message: str, ctx: TraceContext) -> str:
    result = await do_work(user_message)
    return result  # wrapper auto-captures output and closes the span

myAgent = agent("my-agent", run_agent)

# Named access (recommended)
res = await myAgent("hello", {"thread_id": "thread_123"})
print(res.result, res.run_id)

# Tuple unpacking also works
result, run_id, span = await myAgent("hello")

3. Add typed decorators (optional)

Use @tool, @retrieval, @llm, and @trace to add child spans to internal functions:

from uselemma_tracing import agent, tool, retrieval, TraceContext

@retrieval("vector-search")
async def search(query: str) -> list:
    return await vector_db.search(query, top_k=5)

@tool("lookup-order")
async def lookup_order(order_id: str) -> dict:
    return await db.orders.get(order_id)

async def run_agent(user_message: str, ctx: TraceContext) -> str:
    docs = await search(user_message)      # retrieval.vector-search span
    context = "\n".join(docs)
    return await generate_answer(context, user_message)

agent = agent("rag-agent", run_agent)

4. Context manager (no function extraction needed)

from uselemma_tracing import agent

async def handle_request(user_message: str) -> str:
    async with agent("my-agent", input=user_message) as run:
        response = await call_llm(user_message)
        run.complete(response)  # sets ai.agent.output; span closes on block exit
    return response

Export Behavior

• Spans are exported in run-specific batches keyed by lemma.run_id.
• A run batch is exported when its top-level ai.agent.run span ends.
• force_flush() exports remaining runs in separate batches per run.

Environment Variables

Variable	Description
LEMMA_API_KEY	Your Lemma API key
LEMMA_PROJECT_ID	Your Lemma project ID

Both are required unless passed explicitly to register_otel().

Documentation

• Quickstart — first trace in 2 minutes
• Tracing Overview — concepts, API reference, and usage patterns
• OpenAI SDK (Python)
• Anthropic SDK (Python)
• LangChain

License

MIT