This package provides NeedleDocumentStore
and NeedleEmbeddingRetriever
component for use in Haystack projects.
Get started by installing the package via pip
.
pip install needle-haystack-ai
We will show you building a common RAG pipeline using Needle tools and OpenAI generator.
For using these tools you must set your environment variables, NEEDLE_API_KEY
and OPENAI_API_KEY
respectively.
You can get your Needle API key from from Developer settings.
In Needle document stores are called collections. For detailed information, see our docs.
You can create a reference to your Needle collection using NeedleDocumentStore
and use NeedleEmbeddingRetriever
to retrieve documents from it.
from needle_haystack import NeedleDocumentStore, NeedleEmbeddingRetriever
document_store = NeedleDocumentStore(collection_id="<your-collection-id>")
retriever = NeedleEmbeddingRetriever(document_store=document_store)
Use the retriever in a Haystack pipeline. Example:
from haystack import Pipeline
from haystack.components.generators import OpenAIGenerator
from haystack.components.builders import PromptBuilder
prompt_template = """
Given the following retrieved documents, generate a concise and informative answer to the query:
Query: {{query}}
Documents:
{% for doc in documents %}
{{ doc.content }}
{% endfor %}
Answer:
"""
prompt_builder = PromptBuilder(template=prompt_template)
llm = OpenAIGenerator()
# Add components to pipeline
pipeline = Pipeline()
pipeline.add_component("retriever", retriever)
pipeline.add_component("prompt_builder", prompt_builder)
pipeline.add_component("llm", llm)
# Connect the components
pipeline.connect("retriever", "prompt_builder.documents")
pipeline.connect("prompt_builder", "llm")
Run your RAG pipeline:
prompt = "What is the topic of the news?"
result = basic_rag_pipeline.run({
"retriever": {"text": prompt},
"prompt_builder": {"query": prompt}
})
# Print final answer
print(result['llm']['replies'][0])
For detailed guides, take a look at our docs. If you have questions or requests you can contact us in our Discord channel.
needle-haystack
is distributed under the terms of the MIT license.