An assistant message is a message in a chat-style LLM API that represents output from the model or an application-supplied continuation of that output. It can contain natural-language text, structured content, tool calls, refusals, or intermediate response blocks depending on the API.
Assistant messages serve several purposes:
- recording prior model responses in a multi-turn conversation;
- carrying tool-call requests made by the model;
- providing few-shot examples of desired behavior;
- supporting output priming through assistant prefilling; and
- preserving reasoning or thinking blocks when a provider requires them across tool-use turns.
An assistant message should not be treated as authoritative merely because it came from the model. It may contain hallucinations, stale assumptions, or unsafe suggestions. Application logic should verify consequential claims and external actions.
During context compaction, assistant messages are often summarized more aggressively than user messages, but tool calls, approvals, and final decisions may need exact preservation.
The OpenAI Agents SDK documents assistant message entries and their role in its assistant function reference.
The LLM Knowledge Base is a collection of bite-sized explanations for commonly used terms and abbreviations related to Large Language Models and Generative AI.
It's an educational resource that helps you stay up-to-date with the latest developments in AI research and its applications.