Prerequisites

To learn about how pipelines and components work in deepset Cloud, see Pipeline Components and Pipelines.
To use a hosted model, Connect to Model Providers first so that you don't have to pass the API key within the pipeline. For Hugging Face, this is only required for private models. Once deepset Cloud is connected to a model provider, just pass the model name in the model parameter of the component that uses it in the pipeline. deepset Cloud will download and load the model. For more information, see Language Models in deepset Cloud.

An API Key to connect to deepset Cloud.
A ready pipeline to upload to deepset Cloud in the YAML format.
Tip: You can first build your pipeline in Pipeline Builder and then export it to YAML from there.

Pipeline YAML Explained

Your pipeline definition file is in the YAML format. Make sure that you follow the same indentation structure as in this example. Check the Indexing Pipeline and the Query Pipeline format in the tabs:

components: # This section defines your pipeline components and their settings
  component_1: # Give your component a friendly name, you'll use it in the connections section
    type: # You can find the component type in documentation on a component's page
    init_parameters: # Customize the component's settings. To use default values, skip this.
    
  component_2:
    type: 
    init_parameters:
      parameter1: value
      parameter2: value2

# Continue until you define all components
 
connections:  # Define how the components are connected
              # You must explicitly indicate the intpus and outputs you want to connect
              # Input and output types must be the same to be connected
              # You can check components outputs and inputs in documentation
- sender: component_1.output_name # Here you define the output name this component sends to the receiver component
  receiver: component_2.input_name # Here you define the input name that receives the output of the sender component

inputs:  # List all components that need query and filters as inputs but aren't getting them from any other component connected to them
  query:  # These components will receive the query as input
  - "component_1.question"
  filters:  # These components will receive a potential query filter as input
  - "component_1.filters"

components: # This section defines your pipeline components and their settings
  component_1: # Give your component a friendly name, you'll use it in the sections below
    type: # You can find the component type in documentation on a component's page (here maybe a link to components' docs)
    init_parameters: # Customize the component's settings, to use default values, skip this
    
  component_2:
    type: 
    init_parameters:
      parameter1: value1
      parameter2: value2
 
connections:  # Define how the components are connected
              # You must explicitly indicate the intpus and outputs you want to connect
              # Input and output types must be the same to be connected
              # You can check components outputs and inputs in documentation
- sender: component_1.output_name # Here you define the output name this component sends to the receiver component
  receiver: component_2.input_name # Here you define the input name that receives the output of the sender component

inputs:  # List all components that need query and filters as inputs but aren't getting them from any other component connected to them
  query:  # These components will receive the query as input
  - "component_1.question"
  filters:  # These components will receive a potential query filter as input
  - "component_1.filters"
outputs:  # Defines the output of your pipeline, usually the output of the last component
  documents: "component_2.documents"  # The output of the pipeline is the retrieved documents

An example pipeline:

components:
  file_classifier:
    type: haystack.components.routers.file_type_router.FileTypeRouter
    init_parameters:
      mime_types:
      - text/plain
      - application/pdf
      - text/markdown
      - text/html

  text_converter:
    type: haystack.components.converters.txt.TextFileToDocument
    init_parameters:
      encoding: utf-8

  pdf_converter:
    type: haystack.components.converters.pypdf.PyPDFToDocument
    init_parameters:
      converter_name: default

  markdown_converter:
    type: haystack.components.converters.markdown.MarkdownToDocument
    init_parameters:
      table_to_single_line: false

  html_converter:
    type: haystack.components.converters.html.HTMLToDocument
    init_parameters:
      # A dictionary of keyword arguments to customize how you want to extract content from your HTML files.
      # For the full list of available arguments, see
      # the [Trafilatura documentation](https://trafilatura.readthedocs.io/en/latest/corefunctions.html#extract).
      extraction_kwargs:
        output_format: txt  # Extract text from HTML. You can also also choose "markdown"
        target_language: null  # You can define a language (using the ISO 639-1 format) to discard documents that don't match that language.
        include_tables: true  # If true, includes tables in the output
        include_links: false  # If true, keeps links along with their targets

  joiner:
    type: haystack.components.joiners.document_joiner.DocumentJoiner
    init_parameters:
      join_mode: concatenate
      sort_by_score: false

  splitter:
    type: haystack.components.preprocessors.document_splitter.DocumentSplitter
    init_parameters:
      split_by: word
      split_length: 250
      split_overlap: 30

  document_embedder:
    type: haystack.components.embedders.sentence_transformers_document_embedder.SentenceTransformersDocumentEmbedder
    init_parameters:
      model: "intfloat/e5-base-v2"
      device: null

  writer:
    type: haystack.components.writers.document_writer.DocumentWriter
    init_parameters:
      document_store:
        type: haystack_integrations.document_stores.opensearch.document_store.OpenSearchDocumentStore
        init_parameters:
          embedding_dim: 768
          similarity: cosine
      policy: OVERWRITE

connections:  # Defines how the components are connected
- sender: file_classifier.text/plain
  receiver: text_converter.sources
- sender: file_classifier.application/pdf
  receiver: pdf_converter.sources
- sender: file_classifier.text/markdown
  receiver: markdown_converter.sources
- sender: file_classifier.text/html
  receiver: html_converter.sources
- sender: text_converter.documents
  receiver: joiner.documents
- sender: pdf_converter.documents
  receiver: joiner.documents
- sender: markdown_converter.documents
  receiver: joiner.documents
- sender: html_converter.documents
  receiver: joiner.documents
- sender: joiner.documents
  receiver: splitter.documents
- sender: splitter.documents
  receiver: document_embedder.documents
- sender: document_embedder.documents
  receiver: writer.documents

max_loops_allowed: 100

inputs:  # Define the inputs for your pipeline
  files: "file_classifier.sources"  # This component will receive the files to index as input

components:
  chat_summary_prompt_builder:
    type: haystack.components.builders.prompt_builder.PromptBuilder
    init_parameters:
      template: |-
        Rewrite the current question so that it is suitable for web search ONLY if chat history is provided.
        If the chat history is empty, DO NOT reformulate the question. 
        Be cautious when reformulating the current question. Substantial changes to the current question that distort the meaning of the current question are undesirable. 
        It is possible that the current question does not need any changes. 
        The chat history can help to incorporate context into the reformulated question. 
        Make sure to incorporate that chat history into the reformulated question ONLY if needed. 
        The overall meaning of the reformulated question must remain the same as the current question. 
        You cannot change or dismiss keywords in the current question. 
        If you do not want to make changes, just output the current question. 
        Chat History: {{question}}
        Reformulated Question:

  chat_summary_llm:
    type: haystack.components.generators.openai.OpenAIGenerator
    init_parameters:
      api_key: {"type": "env_var", "env_vars": ["OPENAI_API_KEY"], "strict": False}
      model: "gpt-3.5-turbo"
      generation_kwargs:
        max_tokens: 650
        temperature: 0.0
        seed: 0

  replies_to_query:
    type: haystack.components.converters.output_adapter.OutputAdapter
    init_parameters:
      template: "{{ replies[0] }}"
      output_type: str

  bm25_retriever: # Selects the most similar documents from the document store
    type: haystack_integrations.components.retrievers.opensearch.bm25_retriever.OpenSearchBM25Retriever
    init_parameters:
      document_store:
        init_parameters:
          use_ssl: True
          verify_certs: False
          hosts:
            - ${OPENSEARCH_HOST}
          http_auth:
            - "${OPENSEARCH_USER}"
            - "${OPENSEARCH_PASSWORD}"
        type: haystack_integrations.document_stores.opensearch.document_store.OpenSearchDocumentStore
      top_k: 20 # The number of results to return

  query_embedder:
    type: haystack.components.embedders.sentence_transformers_text_embedder.SentenceTransformersTextEmbedder
    init_parameters:
      model: "intfloat/e5-base-v2"
      device: null

  embedding_retriever: # Selects the most similar documents from the document store
    type: haystack_integrations.components.retrievers.opensearch.embedding_retriever.OpenSearchEmbeddingRetriever
    init_parameters:
      document_store:
        init_parameters:
          use_ssl: True
          verify_certs: False
          http_auth:
            - "${OPENSEARCH_USER}"
            - "${OPENSEARCH_PASSWORD}"
        type: haystack_integrations.document_stores.opensearch.document_store.OpenSearchDocumentStore
      top_k: 20 # The number of results to return

  document_joiner:
    type: haystack.components.joiners.document_joiner.DocumentJoiner
    init_parameters:
      join_mode: concatenate

  ranker:
    type: haystack.components.rankers.transformers_similarity.TransformersSimilarityRanker
    init_parameters:
      model: "intfloat/simlm-msmarco-reranker"
      top_k: 8
      device: null
      model_kwargs:
        torch_dtype: "torch.float16"

  qa_prompt_builder:
    type: haystack.components.builders.prompt_builder.PromptBuilder
    init_parameters:
      template: |-
        You are a technical expert.
        You answer questions truthfully based on provided documents.
        For each document check whether it is related to the question.
        Only use documents that are related to the question to answer it.
        Ignore documents that are not related to the question.
        If the answer exists in several documents, summarize them.
        Only answer based on the documents provided. Don't make things up.
        If the documents can't answer the question or you are unsure say: 'The answer can't be found in the text'.
        These are the documents:
        {% for document in documents %}
        Document[{{ loop.index }}]:
        {{ document.content }}
        {% endfor %}
        Question: {{question}}
        Answer:

  qa_llm:
    type: haystack.components.generators.openai.OpenAIGenerator
    init_parameters:
      api_key: {"type": "env_var", "env_vars": ["OPENAI_API_KEY"], "strict": False}
      model: "gpt-3.5-turbo"
      generation_kwargs:
        max_tokens: 400
        temperature: 0.0
        seed: 0

  answer_builder:
    init_parameters: {}
    type: haystack.components.builders.answer_builder.AnswerBuilder

connections:  # Defines how the components are connected
- sender: chat_summary_prompt_builder.prompt
  receiver: chat_summary_llm.prompt
- sender: chat_summary_llm.replies
  receiver: replies_to_query.replies
- sender: replies_to_query.output
  receiver: bm25_retriever.query
- sender: replies_to_query.output
  receiver: query_embedder.text
- sender: replies_to_query.output
  receiver: ranker.query
- sender: replies_to_query.output
  receiver: qa_prompt_builder.question
- sender: replies_to_query.output
  receiver: answer_builder.query
- sender: bm25_retriever.documents
  receiver: document_joiner.documents
- sender: query_embedder.embedding
  receiver: embedding_retriever.query_embedding
- sender: embedding_retriever.documents
  receiver: document_joiner.documents
- sender: document_joiner.documents
  receiver: ranker.documents
- sender: ranker.documents
  receiver: qa_prompt_builder.documents
- sender: ranker.documents
  receiver: answer_builder.documents
- sender: qa_prompt_builder.prompt
  receiver: qa_llm.prompt
- sender: qa_llm.replies
  receiver: answer_builder.replies

max_loops_allowed: 100

inputs:  # Define the inputs for your pipeline
  query:  # These components will receive the query as input
  - "chat_summary_prompt_builder.question"

  filters:  # These components will receive a potential query filter as input
  - "bm25_retriever.filters"
  - "embedding_retriever.filters"

outputs:  # Defines the output of your pipeline
  documents: "ranker.documents"  # The output of the pipeline is the retrieved documents
  answers: "answer_builder.answers"  # The output of the pipeline is the retrieved documents

Create a Pipeline

Follow the step-by-step code explanation:

▶️

Create a Pipeline with API

Open Recipe

Or use the following code:

curl --request POST \
     --url https://api.cloud.deepset.ai/api/v1/workspaces/<YOUR_WORKSPACE>/pipelines \
     --header 'Accept: application/json' \
     --header 'Authorization: Bearer <YOUR_API_KEY>'\
     --data-binary "@path/to/pipeline.yaml"

In line 2, replace <YOUR_WORKSPACE> with the name of the workspace where you want to create the pipeline.
In line 4, replace <YOUR_API_KEY> with the deepset Cloud API key.
In line 5, replace "@path/to/pipeline.yaml" with the path to your pipeline YAML file.

See the REST API endpoint documentation.