Haystack

Trace Deepset Haystack pipelines with W&B Weave using the WeaveConnector integration.

Haystack is an open source framework for building search and LLM applications. Deepset maintains a WeaveConnector component that forwards Haystack pipeline traces to W&B Weave so you can inspect component runs, prompts, and outputs in the Weave UI.

For full API details and additional examples, see these Deepset resources:

• Haystack’s WeaveConnector.
• Haystack’s Weave integration API reference.
• Haystack’s Trace with W&B Weave example using a RAG pipeline.

Prerequisites

Before you begin:

• Set WANDB_API_KEY in your environment using your W&B API key.
• Set HAYSTACK_CONTENT_TRACING_ENABLED to true before you run a pipeline so Haystack emits tracing data the connector can forward.

Install

Install the required dependencies using pip:

pip install weave-haystack

The package declares compatible versions of haystack-ai and weave as dependencies.

Trace a Haystack pipeline with Weave

The following example adds Haystack’s WeaveConnector to a Haystack Pipeline and integrates with W&B Weave for tracing and monitoring your pipeline components. The pipeline_name you pass is used as the Weave project name for traces from that pipeline.

In your Haystack pipeline, don’t connect WeaveConnector to other components.

import os

os.environ["HAYSTACK_CONTENT_TRACING_ENABLED"] = "true"

from haystack import Pipeline
from haystack.components.builders import ChatPromptBuilder
from haystack.components.generators.chat import OpenAIChatGenerator
from haystack.dataclasses import ChatMessage
from haystack_integrations.components.connectors.weave import WeaveConnector

pipe = Pipeline()
pipe.add_component("prompt_builder", ChatPromptBuilder())
pipe.add_component("llm", OpenAIChatGenerator(model="gpt-3.5-turbo"))
pipe.connect("prompt_builder.prompt", "llm.messages")

# pipeline_name becomes your W&B project name.
connector = WeaveConnector(pipeline_name="haystack_demo")
# Add connector to pipeline but don't connect it.
pipe.add_component("weave", connector)

messages = [
    ChatMessage.from_system(
        "Always respond in German even if some input data is in other languages.",
    ),
    ChatMessage.from_user("Tell me about {{location}}"),
]

response = pipe.run(
    data={
        "prompt_builder": {
            "template_variables": {"location": "Berlin"},
            "template": messages,
        },
    },
)

print(response["llm"]["replies"][0])

After the pipeline runs, open your W&B workspace, select the project named with pipeline_name, and go to Traces to review the completed trace.