Skip to main content

​
Signature

def instrument_baml(baml_client: Any) -> Any

​
Overview

Unlike Anthropic, OpenAI, LangChain, and Gemini — which are auto-instrumented when init() is called — BAML clients are code-generated per project and cannot be monkey-patched. You must explicitly wrap the client with instrument_baml() after calling init().
If you use BAML without calling instrument_baml(), calls will not be traced, even if init() was called. There is no error — they simply won’t appear in traces.

​
Parameters

​
baml_client
Any
required
The BAML client instance to wrap. Typically imported as b from baml_client.

​
Returns

A proxy-wrapped client with tracing enabled. Assign the result back to your client variable.

​
Example

from baml_client import b
from rdk import init, observe, instrument_baml

init(
    endpoint="https://collector.021labs.ai",
    api_key="your-api-key",
)

# Required: wrap the BAML client after init()
b = instrument_baml(b)

@observe(name="extract-entities")
async def process(text: str):
    return await b.ExtractEntities(text)  # Auto-traced!

​
How It Works

instrument_baml() wraps the BAML client in a proxy that intercepts method calls. When a BAML function is called:
  1. The proxy detects the active trace (from @observe or trace())
  2. A span is created with the function name and arguments
  3. The real BAML function executes normally
  4. The span is completed with the structured result and token usage

​
What Gets Captured

FieldDescription
typeLLM
nameBAML function name (e.g., baml.ClassifyInquiry)
input.argsFunction arguments
output.resultStructured result
token_usageToken counts
metadata.provider"baml"

​
See Also