LangWatch Observability Integration

This package integrates LangWatch tracing with the Bridgic framework, providing worker granularity tracing implementation.

Installation

# Install the LangWatch tracing package
pip install bridgic-traces-langwatch

Prerequisites

LangWatch provides a hosted version of the platform, or you can run the platform locally.

• To use the hosted version, you need to create a LangWatch account and grab your API key from the dashboard.
• To run LangWatch locally, see the self-hosting guide for more information.

Configuration

Using Environment Variables

Set the following environment variables:

export LANGWATCH_API_KEY="your-api-key-here"
export LANGWATCH_ENDPOINT="https://app.langwatch.ai"  # Optional, defaults to https://app.langwatch.ai

Usage

The LangWatchTraceCallback can be configured in two ways:

Method 1: Per-Automa Scope with RunningOptions

Apply the callback only to a single automa by configuring it through RunningOptions. In this mode, every worker instantiated by that automa receives its own callback instance, while other automa remain unaffected.

from bridgic.core.automa import GraphAutoma, RunningOptions, worker
from bridgic.core.automa.worker import WorkerCallbackBuilder
from bridgic.traces.langwatch import LangWatchTraceCallback
import asyncio

class MyAutoma(GraphAutoma):
    @worker(is_start=True)
    async def step1(self):
        await asyncio.sleep(1)
        return "hello"

    @worker(dependencies=["step1"], is_output=True)
    async def step2(self, step1: str):
        await asyncio.sleep(1)
        return f"{step1} world"

async def main():
    builder = WorkerCallbackBuilder(
        LangWatchTraceCallback,
        init_kwargs={"base_attributes": {"app": "demo"}}
    )
    running_options = RunningOptions(callback_builders=[builder])
    automa = MyAutoma(running_options=running_options)
    result = await automa.arun()
    print(result)

if __name__ == "__main__":
    asyncio.run(main())

Method 2: Global Scope with GlobalSetting

You can register the callback at the global level through GlobalSetting to make it effective for every automa in the runtime. Each worker, regardless of which automa creates it, is instrumented with the same callback configuration.

from bridgic.core.automa import GraphAutoma, worker
from bridgic.core.automa.worker import WorkerCallbackBuilder
from bridgic.core.config import GlobalSetting
from bridgic.traces.langwatch import LangWatchTraceCallback
import asyncio

# Configure global callback
GlobalSetting.set(callback_builders=[WorkerCallbackBuilder(
    LangWatchTraceCallback,
    init_kwargs={"base_attributes": {"app": "demo"}}
)])

class DataAnalysisAutoma(GraphAutoma):
    @worker(is_start=True)
    async def collect_data(self, topic: str) -> dict:
        """Collect data for the given topic."""
        # Simulate data collection
        await asyncio.sleep(1)
        return {
            "topic": topic,
            "data_points": ["point1", "point2", "point3"],
            "timestamp": "2024-01-01"
        }

    @worker(dependencies=["collect_data"])
    async def analyze_trends(self, data: dict) -> dict:
        """Analyze trends in the collected data."""
        # Simulate trend analysis
        await asyncio.sleep(1)
        return {
            "trends": ["trend1", "trend2"],
            "confidence": 0.85,
            "source_data": data
        }

    @worker(dependencies=["analyze_trends"], is_output=True)
    async def generate_report(self, analysis: dict) -> str:
        """Generate a final report."""
        await asyncio.sleep(1)
        return f"Report: Found {len(analysis['trends'])} trends with {analysis['confidence']} confidence."

async def main():
    automa = DataAnalysisAutoma()
    result = await automa.arun(topic="market analysis")
    print(result)

if __name__ == "__main__":
    asyncio.run(main())

From the perspective of tracking worker execution, the following approach is equivalent to the above one:

from bridgic.traces.langwatch import start_langwatch_trace

start_langwatch_trace(base_attributes={"app": "demo"})

However, start_langwatch_trace is a higher-level function that encapsulates the functionality of the first approach. As the framework may add tracking for more important phases in the future, start_langwatch_trace will provide a unified interface for all tracking capabilities, making it the recommended approach for most use cases. When parameters are omitted, the helper reads the values from the LANGWATCH_API_KEY and LANGWATCH_ENDPOINT environment variables.