# Large Language Model Operations (LLMOps)

> How teams run apps built on language models. It covers prompts, tests and cost.

**Field:** Artificial Intelligence  
**Category:** MLOps & Deployment  
**Also known as:** LLMOps, LLM Ops, GenAI Ops

## Think of it like

It is like running a shop, not making one cake. You keep it all going each day.

## Definition

LLMOps is the work of running apps built on large language models. It grew out of MLOps, but most teams do not train their own model. Instead they manage prompts, pick and swap models, and test answers with evals. They also trace each request, watch cost and speed, and add safety checks. The goal is an app that stays good and safe as models and users change.

## Example

Jack's team runs a help chatbot. Each week they test new prompts on 200 saved questions, track cost per chat and roll back any change that makes answers worse.

## Why it matters

A chatbot demo is easy. Keeping it right, safe and cheap for thousands of users is the hard part.

## Common mistake

Changing a prompt with no test set. A fix for one question can quietly break ten others.

## Three principles

- **Prompts are code.** Version them, test them and roll them back.
- **Test with evals.** Saved questions catch a change that makes answers worse.
- **Watch every call.** Traces show cost, speed and failures.

## Related terms

- [MLOps](https://learn.benibix.com/term/ai/mlops)
- [Prompt Versioning](https://learn.benibix.com/term/development/prompt-versioning)
- [LLM Observability](https://learn.benibix.com/term/development/llm-observability)
- [LLM Evaluation](https://learn.benibix.com/term/ai/llm-evaluation)
- [LLM Gateway](https://learn.benibix.com/term/development/llm-gateway)
- [Guardrail (AI)](https://learn.benibix.com/term/ai/guardrail)

## Told apart from

- [MLOps vs LLMOps](https://learn.benibix.com/compare/mlops-vs-llmops): People think LLMOps is a new field that replaces MLOps, and it is not. It is the same habit for apps built on a language model. The loop of test, ship and watch stays. What changes is the part you edit. In MLOps you retrain the model. In LLMOps you rarely do, so you version the prompt, test it with saved questions and trace each call.

![A loop of five boxes: data, train, track runs, registry, serve and watch, then back to data. AutoML sits on the train box, MLflow and Weights and Biases on the track box. A bracket under the loop says Kubeflow runs the steps on Kubernetes. A side box says LLMOps adds prompts, traces and evals.](https://learn.benibix.com/diagrams/ai7-mlops-tools.svg)

---

Source: https://learn.benibix.com/term/ai/llmops
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-17
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
