> For the complete documentation index, see [llms.txt](https://gdplabs.gitbook.io/glchat/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://gdplabs.gitbook.io/glchat/developer-documentation/glchat-pipeline.md).

# GLChat Pipeline

> This section explains how GLChat processes a user message end-to-end — from the moment it arrives to the moment a response is streamed back

## Overview

When a user sends a message, it does not go directly to an AI model. It passes through a series of stages that handle input processing, safety, context enrichment, execution, and persistence. Understanding this flow helps you know exactly where things happen and why.

### Full Execution Flow

<figure><img src="/files/VqYOpfgHoih3Wg5Hnd6e" alt=""><figcaption></figcaption></figure>

### Stage Summary

#### Guardrails

**Guardrails** runs a safety check on the raw user input. If the input violates configured policies, a blocked response is returned immediately. If guardrails are not enabled or the input is safe, the message continues to preprocessing.

#### Preprocessing

Preprocessing enriches the raw message with full conversation context — rewriting it as a standalone query, retrieving long-term memories, and checking the response cache. If a cache hit is found, the cached response is returned immediately without running the pipeline.

#### Routing

After preprocessing, the system revisits the routing decision made at the very start. It now applies it: the message is directed to a **Pipeline** (e.g. Standard RAG) or an **Agent**, depending on the user configuration.

#### Pipeline or Agent

* **Pipeline** — a structured, deterministic flow that retrieves context and generates a response. Standard RAG is the primary pipeline.
* **Agent** — a flexible, multi-step executor that can call tools, reason, and loop before producing a final answer.

Both paths produce a response that is streamed to the user in real time.

#### Postprocessing

After the pipeline or agent finishes, the system saves the interaction — chat history and memory — asynchronously so it does not delay the response stream.