About This Task

What Are Custom Components

A component is a Python code snippet that follows our template and performs a specific task on your data. When you create and upload a custom component to deepset AI Platform, it becomes accessible to your entire organization. Any member can then use the component in their pipelines.

Custom components are based on Haystack components. Haystack is deepset's open source AI framework, which also powers deepset AI Platform. To learn more, visit the Haystack website.

Custom Components Template

We provide a template for creating your custom components, which are available as a GitHub repository. This template serves as a custom components library for your organization. Components created in the ./dc-custom-component-template/src/dc_custom_component/example_components/ folder and imported into deepset AI Platform are the components you can use in your pipelines. Only the components present in the most recently uploaded template are available for use.

For example, if someone in your organization creates a component called WelcomeTextGenerator and uploads it to deepset, everyone in the organization can use it. However, if later someone adds two new components, GoodbyeTextGenerator and CharacterSplitter, and deletes WelcomeTextGenerator, only the new components will be available to use in your pipelines. WelcomeTextGenerator will no longer be accessible.

Currently, you can't delete custom components.

Custom Components in Pipeline Builder

Custom components uploaded to deepset AI Platform are shown in the Custom tab in Pipeline Builder. They're grouped automatically based on their path in the custom components package.

For example, if your component is located at/dc-custom-component-template/src/dc_custom_component/components/generators in the package, it will show up under the Generators group within the Custom tab.

Prerequisites

Read the following resources before completing this task:
You should have a basic understanding of working with GitHub repositories.
Generate a deepset API key to upload the created component to your workspace. For instructions, see Generate an API Key.

Create a Component

Prepare the Template

Fork the dc-custom-component-template GitHub repository. This will let you version control your changes.
Navigate to the directory where you cloned the repository and open the ./dc-custom-component-template/src/dc_custom_component/example_components/ directory. The preprocessors and rankers folders are examples you can modify or remove as needed.
Create a new folder or rename one of the example folders to match your custom component's name and open it. There's a .py file inside. This is where you'll write your component code. You can rename this file as well.
Example:
To create a custom component called WelcomeTextGenerator:
1. Rename ./dc-custom-component-template/src/dc_custom_component/example_components/preprocessors to ./dc-custom-component-template/src/dc_custom_component/components/generators.
2. Open the generators folder and rename the example file character_splitter.py to welcome_text_generator.py.
3. Delete the rankers folder if it's not needed. If you're creating multiple components, use the folder structure to keep them organized.
  Note: You can create all your components in one .py file or you can have a separate folder with a .py file in it for each custom component. That's up to you.
  📘
  Component Folder as Group Name
  The folder name containing your component code becomes the component group name in Pipeline Builder. For example, if you place your component in ./dc-custom-component-template/src/dc_custom_component/components/generators, it will appear in the Generators group in the Pipeline Builder component library within the Custom tab.

Set Up a Virtual Environment

Creating a virtual environment isolates your project's dependencies from other Python projects and your system Python installation.

The template uses Hatch, a Python project manager, to set up virtual environments. For details, see the Hatch website.

Install Hatch by running: pip install hatch. This installs all the necessary packages, including pytest.
Create a virtual environment by running: hatch shell.

Implement the Component

Write the component code in the .py file. Use the recipe below as a starting point:

▶️

Creating a Custom Component

Open Recipe

If your component has dependencies, add them in the ./dc-custom-component-template/pyproject.toml file in the dependencies section:

dependencies = [
  "haystack-ai>=2.0.0"
]

Note: Do not modify versions of dependencies already listed in this file.

From the project root directory, run thehatch run code-quality:all command to format your code.
Update the component version in the ./dc-custom-component-template/src/dc_custom_component/__about__.py file. You can specify version numbers in any way you like, but we suggest you adopt the major.minor.micro format, for example, 1.1.0. Have a look at Hatch versioning for guidelines.
The version number applies to all components. Even if you have multiple components, you only need to specify one version number.

Components Connecting to Third-Party Providers

If your component connects to a third-party provider that requires authentication, store the API key in an environment variable and then retrieve it in the component's run() method. You can also add a secret on the Secrets page and give it the same name as the environment variable. For details, see Add Secrets to Connect to Third Party Providers.

The following component shows how to do this. First, you add the env_var_name init parameter and then you retrieve this value in the run() method using os.getenv(self.env_var_name). The retrieved value is stored in loaded_api_key.

@component
class CustomComponent:
    def __init__(self, env_var_name: str = "API_KEY"):
        self.env_var_name = env_var_name

    @component.output_types(answers=List[GeneratedAnswer])
    def run(self, query: str) -> Dict[str, List[GeneratedAnswer]]:
        loaded_api_key = os.getenv(self.env_var_name)
        # use the key
        pass

Using Secrets for Secure Connections

Use Secrets in Your Custom Components

Add the secret to your component's code:

Make sure you import it.
Specify the environment variable name, which must match the secret's name.
Add its serialization and deserialization methods.
Add a warm_up() method to load the API key before pipeline validation. This step is optional but we recommend it:

from haystack import component, default_from_dict, default_to_dict
from haystack.utils import Secret, deserialize_secrets_inplace
from typing import Any, Dict


@component
class MyComponent:
    def __init__(self, model_name: str, api_key: Secret = Secret.from_env_var("ENV_VAR_NAME")):
        self.model_name = model_name
        self.api_key = api_key
        self.backend = None

    def warm_up(self):
        # Call api_key.resolve_value() to load the API key from the environment variable
        # We put the resolution in warm_up() to avoid loading the API key during pipeline validation
        if self.backend is None:
            self.backend = SomeBackend(self.model_name, self.api_key.resolve_value())

    def to_dict(self) -> Dict[str, Any]:
        # Make sure to include any other init parameters in the to_dict method
        return default_to_dict(
            self,
            model_name=self.model_name,
            api_key=self.api_key.to_dict(),
        )

    @classmethod
    def from_dict(cls, data: Dict[str, Any]) -> "MyComponent":
        # Make sure to use deserialize_secrets_inplace to load the Secret object
        init_params = data.get("init_parameters", {})
        deserialize_secrets_inplace(init_params, keys=["api_key"])
        return default_from_dict(cls, data)

    def run(self, my_input: Any):
        if self.backend is None:
            raise RuntimeError("The component wasn't warmed up. Run 'warm_up()' before calling 'run()'.")
        return self.backend.process(my_input)

If you have multiple secrets for one provider, you can easily switch between them in your pipeline YAML by updating the secret's name:

llm:
  type: dc_custom_component.components.my_components.component1.MyComponent # the path to your component
  init_parameters:
     api_key: {"type": "env_var", "env_vars": ["ENV_VAR_NAME"], "strict": False} # uses the `ENV_VAR_NAME` secret
     # to use another secret, update its name here

Test Your Component

We recommend that you test your component before uploading.

Add unit and integration tests in the ./dc-custom-component-template/tests folder to ensure everything works fine.
Run your tests using: hatch run tests. If the tests pass, your component is ready.

What to Do Next

Upload your component to deepset AI Platform to use it in your pipelines.