How to Add Custom Deep Research Pipeline to GLChat
You can add your own custom Deep Research pipeline to GLChat. This allows you to use a different research provider, integrate custom tools or data sources, or apply specialized logic that the built-in profiles (Essentials, Comprehensive, Internal) do not offer.
The process has two main steps: create a Deep Research pipeline, then register it with GLChat using the existing pipeline registration flow.
Step 1: Create a Deep Research Pipeline
A Deep Research pipeline is a pipeline that uses the GL SDK DeepResearcher component as one of its steps. The DeepResearcher is a language binding that delegates to an external deep research provider (e.g., OpenAI, Parallel.ai, Perplexity, Google). You choose which provider to use, and can optionally configure tools (e.g., MCP, web search) that the provider supports.
What You Need
GL SDK DeepResearcher — The core component that executes deep research. You place this component in your pipeline as a step. See GL SDK DeepResearcher for the component documentation and available providers.
Pipeline structure — The GL SDK DeepResearcher component is end-to-end deep research by itself. You can add pre and post steps around it. For example, a router before DeepResearcher can decide, based on the user query, whether to invoke deep research or use a cheaper path (e.g., normal generation) to save cost. For how to build a pipeline in general, see Your First RAG Pipeline.
Full example — For a complete example that shows a Deep Research pipeline with routing (deciding when to invoke deep research vs. normal generation), see Deep Research Pipeline with Routing.
Summary
To create a Deep Research pipeline, you put the GL SDK DeepResearcher component in your pipeline. You can add routing logic, custom prompts, MCP tools, or other steps around it as needed for your use case.
Step 2: Add the Pipeline to GLChat
Once your Deep Research pipeline is built, you register it with GLChat so it becomes available as a new application for users. GLChat already supports adding custom pipelines—you use the same registration flow as for other custom pipelines (e.g., RAG).
See How to add a pipeline to GLChat for details on the registration process.
Last updated