Quick Start

Create your first custom pipeline in under 5 minutes.

⚠️ Disclaimer: This guide is specifically for adding new custom pipelines within the GLChat. The instructions and examples provided here are tailored to GLChat's architecture and infrastructure. They are not intended for use in other projects or standalone applications.

Prerequisites

Before creating your custom pipeline, ensure you have completed the following setup:

1. Setup GLChat Backend

Follow the setup instructions in the GLChat Backend README to:

• Install dependencies

• Configure environment variables

• Set up the database

• Start the required services

2. Get OpenAI API Key

1. Sign up or log in to OpenAI Platform

2. Navigate to API Keys

3. Create a new API key

4. Add the API key to your .env file:

OPENAI_API_KEY=your-api-key-here

Note: Replace your-api-key-here with your actual OpenAI API key. Keep your API key secure and never commit it to version control.

Step 1: Generate Pipeline Files

Use the automated CLI tool to create your pipeline structure:

make new-pipeline

Or directly:

uv run python scripts/create_pipeline.py

The script will prompt you for:

• Pipeline Name: Lowercase, alphanumeric with underscores (e.g., my_custom_pipeline)

• Description: Brief description of the pipeline's purpose

• Author Name: Your full name

• Author Email: Your email (should be @gdplabs.id)

Naming Convention:

• Folder Name: Use snake_case (e.g., my_custom_pipeline)

• Pipeline ID: System converts to kebab-case (folder: my_custom_pipeline → ID: my-custom-pipeline)

The CLI tool automatically:

• Creates the pipeline folder: glchat_be/config/pipeline/<pipeline_name>/

• Generates __init__.py and pipeline.py with proper stubs and docstrings

• Creates a migration file with auto-detected down_revision

• Provides next steps guidance

Example Session

$ make new-pipeline

GLChat Pipeline Generator

Pipeline Name (lowercase, alphanumeric with underscores): my_custom_pipeline
Description: Custom pipeline for processing X data
Author Name: John Doe
Author Email: [e-mail]

Summary:
  Pipeline Name: my_custom_pipeline
  Description: Custom pipeline for processing X data
  Author: John Doe ([e-mail])

Files to be created:
  1. glchat_be/config/pipeline/my_custom_pipeline/
     └── __init__.py
     └── pipeline.py
  2. migration/versions/v310_add_my_custom_pipeline_pipeline.py

Create pipeline? [Y/n]: y

Creating pipeline files...
✓ Created pipeline folder: glchat_be/config/pipeline/my_custom_pipeline
✓ Created __init__.py
✓ Created pipeline.py (stub)

Detecting migration dependencies...
  ✓ Auto-detected down_revision: b5b4fcfe43a7

Creating migration file...
✓ Created migration: migration/versions/v310_add_my_custom_pipeline_pipeline.py

✓ Pipeline created successfully!

For more details about the CLI tool, see scripts/README.md.

Step 2: Run Migration

Apply the database migration to register your pipeline:

make migrate

This registers your pipeline in the database, making it available in the Admin Dashboard.

That's it! Your pipeline is now ready to use. The CLI tool generates a complete, working LLM pipeline that you can use immediately.

What does the generated pipeline look like?

The CLI tool generates a complete pipeline.py file with two main functions:

The build() Function - Constructs the pipeline using GL SDK components:

"""Custom pipeline for processing X data.

Authors:
    John Doe ([e-mail])

References:
    NONE
"""

from typing import Any
import os

from gllm_generation.response_synthesizer import ResponseSynthesizer
from gllm_inference.catalog.catalog import BaseCatalog
from gllm_pipeline.pipeline.pipeline import Pipeline
from gllm_pipeline.steps import step


async def build(
    pipeline_config: dict[str, Any],
    prompt_builder_catalogs: dict[str, BaseCatalog[Any]] | None = None,
    lmrp_catalogs: dict[str, BaseCatalog[Any]] | None = None,
) -> Pipeline:
    """Constructs the pipeline instance.

    Args:
        pipeline_config (dict[str, Any]): Configuration for the pipeline (model, settings, etc.)
        prompt_builder_catalogs (dict[str, BaseCatalog[Any]]): Optional catalogs for prompt templates.
        lmrp_catalogs (dict[str, BaseCatalog[Any]]): Optional catalogs for language model resources.

    Returns:
        Pipeline: A constructed Pipeline object.
    """
    response_synthesizer = ResponseSynthesizer.stuff_preset(
        model_id="openai/gpt-5-mini",
        credentials=os.getenv("OPENAI_API_KEY")
    )

    return Pipeline(steps=[
        step(
            component=response_synthesizer,
            input_map={
                "query": "user_query",
                "history": "history"
            },
            output_state="response"
        )
    ])


def build_initial_state(
    request: dict[str, Any],
    pipeline_config: dict[str, Any],
    previous_state: dict[str, Any] | None = None,
    **kwargs: Any,
) -> dict[str, Any]:
    """Creates the starting state for a pipeline execution.

    Args:
        request (dict[str, Any]): The incoming request payload from the API/User.
        pipeline_config (dict[str, Any]): The pipeline's configuration.
        previous_state (dict[str, Any] | None): The previous state. Defaults to None.
        **kwargs: Additional context.

    Returns:
        A dictionary containing the initial state fields.
    """
    if not previous_state:
        raise ValueError("Previous state is required")

    return {
        "user_query": previous_state.get("user_query"),
        "event_emitter": previous_state.get("event_emitter"),
        "history": previous_state.get("history", []),
    }

The build_initial_state() Function - Prepares the initial context/state that flows through the pipeline.

This is a simple LLM-only pipeline that's ready to use out-of-the-box. For building more complex RAG pipelines with retrievers, rerankers, and other components, refer to the official GL SDK guide.

---

Next Steps

Your pipeline is ready to use! Here are some optional next steps:

1. Customize your pipeline (optional) - The generated pipeline is a simple LLM-only pipeline. To add retrieval, reranking, or other advanced features, modify glchat_be/config/pipeline/<pipeline_name>/pipeline.py. See the official GL SDK guide for examples.

2. Customize the migration (optional) - If you need organization-specific settings or other customizations, modify the generated migration file. See Database Migration Guide

3. Add advanced configuration (optional) - Configure custom settings, parameters, and more. See Advanced Configuration Guide