This section covers how a message actually travels through an agent — from the moment it arrives to the moment an answer is shown — and the settings that shape each step.
Overview
Preprocessing handles the incoming message before the model reasons over it — query rephrasing, chunk retrieval, and ranking — and compacts long conversation history so it stays within the model's context window. Response Generation shapes the outgoing answer — output format, language, tone, and whether the agent's reasoning is shown. Agent Runtime ties it all together as a concept doc: the full execution pipeline, including how Instructions, Tools, Guardrails, and Models are actually assembled and consulted on every reasoning step.
What's in This Section
Topic | Covers |
|---|---|
Preprocessing | Query rephrasing, chunk retrieval and ranking of the incoming message, and compaction of long conversation history. |
Response Generation | Output format, language, tone, chain-of-thought, and watermarking of the outgoing answer. |
Agent Runtime | The full execution pipeline — how every section of the agent editor comes together at request time. |
A Good Reading Order
Read Agent Runtime first if you want the big picture of how a request flows end to end; read Preprocessing and Response Generation when you need to tune one specific stage of that flow.