Pipeline

A pipeline is a structured, deterministic execution path that retrieves relevant context and generates a grounded response. GLChat ships with four main built-in pipelines: Standard RAG, No-Op, Gemini Live, and Deep Research.

Available Pipelines

Standard Pipelines

These are the preset pipelines available to every project configuration. They cover the most common deployment scenarios.

Pipeline
Description
Use case

Standard RAG

Retrieves from a knowledge base, web, or database, then generates a grounded answer

Knowledge base Q&A, document search, web-augmented answers

No-Op

Skips retrieval entirely — the LLM responds without any process beforehand

General conversation when no knowledge base is configured

External Pipeline

Calls a configurable external webhook endpoint using an OpenAI-compatible interface

Connecting GLChat to any external LLM or service via HTTP

Datasaur External

Calls the Datasaur LLM API with built-in citation support and Datasaur-specific reference formatting

Chatbots powered by Datasaur's hosted LLM platform

Specialized Pipelines

These pipelines are not available as preset options. They are activated only when a specific feature flag or search type is present in the request — and each requires a dedicated backend handler to be configured via environment variables.

Pipeline
Activation condition
Use case

Gemini Live

enable_live: true + BACKEND_GEMINI_LIVE_PIPELINE_HANDLER configured

Interactive voice and live conversation scenarios

Deep Research

search_type is ESSENTIALS_DEEP_RESEARCH, COMPREHENSIVE_DEEP_RESEARCH, or INTERNAL_DEEP_RESEARCH + BACKEND_ESSENTIALS_DEEP_RESEARCH_PIPELINE_HANDLER configured

In-depth research queries requiring breadth and synthesis across multiple sources

Want to use or build a custom pipeline?

GLChat supports connecting your own external pipeline or even creating your own custom pipeline. For a step-by-step guide, see Custom Pipeline Development Guide

Last updated