> ## Documentation Index
> Fetch the complete documentation index at: https://docs.galtea.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# start_as_current_observation

> Drop-in replacement for Langfuse's start_as_current_observation context manager with Galtea trace linking.

## Example

```python theme={"system"}
from galtea.integrations.langfuse import start_as_current_observation

# Without Galtea correlation:
with start_as_current_observation(name="process-query", as_type="span") as span:
    result = run_agent("user query")
    span.update(output=result)

# With Galtea correlation:
with start_as_current_observation(
    name="process-query",
    as_type="span",
    inference_result_id="inferenceResult_abc123",
) as span:
    result = run_agent("user query")
    span.update(output=result)
```

<Note>
  This is a synchronous context manager. Inside `async def` code, use plain `with` (not `async with`); `async with` raises `TypeError` because the underlying generator does not implement the async context manager protocol. For async agents, prefer the [`@observe`](/sdk/api/langfuse/observe) decorator.
</Note>

## Parameters

<ResponseField name="inference_result_id" type="str" optional>
  Galtea inference result ID to link traces to. When provided, the wrapper manages the Galtea trace context automatically for the duration of the block.
</ResponseField>

<ResponseField name="name" type="str" required>
  Span name for the observation.
</ResponseField>

<ResponseField name="as_type" type="str" optional>
  Langfuse observation type: `span` (default), `generation`, `agent`, `tool`, `retriever`, `chain`, `evaluator`, `embedding`, `guardrail`. Mapped to Galtea's `TraceType` automatically. See [observation types](/sdk/integrations/langfuse/overview#observation-types).
</ResponseField>

All other keyword arguments are forwarded to Langfuse's `start_as_current_observation`.

## Yields

A Langfuse observation span object. Supports all native Langfuse span methods:

<ResponseField name="update(**kwargs)" type="method">
  Update observation attributes (`output`, `model`, `metadata`, etc.).
</ResponseField>

<ResponseField name="start_as_current_observation(**kwargs)" type="method">
  Create a child observation. Child calls are **native Langfuse** — no Galtea wrapper needed.
</ResponseField>

<ResponseField name="score(**kwargs)" type="method">
  Attach a score to this observation.
</ResponseField>

## Nested Observations

Only the **root** call needs the Galtea wrapper. Child observations created via `root.start_as_current_observation(...)` are native Langfuse — they are automatically linked to Galtea while the root context is active.

```python theme={"system"}
from galtea.integrations.langfuse import start_as_current_observation

# Only the root call needs the Galtea wrapper.
# Child calls on yielded spans are native Langfuse.
with start_as_current_observation(
    name="pipeline",
    as_type="span",
    inference_result_id="inferenceResult_abc123",
) as root:
    with root.start_as_current_observation(name="retrieve", as_type="retriever") as ret:
        docs = ["doc1", "doc2"]
        ret.update(output=docs)

    with root.start_as_current_observation(name="generate", as_type="generation") as gen:
        response = "Generated response"
        gen.update(output=response, model="gpt-4")

    root.update(output=response)
```