---
title: "LLM OS"
description: "An LLM OS is a computational architecture proposed by Andrej Karpathy in a post on X. In contrast to a traditional operating system, it is not deterministic and relies on Large Language Models (LLM) to process instructions."
canonical_url: "https://promptmetheus.com/resources/llm-knowledge-base/llm-os"
---

> Index of every page on this site: [llms.txt](https://promptmetheus.com/llms.txt)

An LLM OS is a computational architecture proposed by [Andrej Karpathy](https://x.com/karpathy) in a [post on X](https://x.com/karpathy/status/1723140519554105733). In contrast to a traditional operating system, it is not deterministic and relies on [Large Language Models (LLM)](https://promptmetheus.com/resources/llm-knowledge-base/large-language-model-llm) to process instructions.

At the core of the architecture is an LLM [inference](https://promptmetheus.com/resources/llm-knowledge-base/inference) engine (analogous to the CPU in a computer), which is connected to other LLMs, peripheral devices (video, audio, etc.), the internet, "tools" (calculator, code interpreter, etc.), and a file system (database with [vector embeddings](https://promptmetheus.com/resources/llm-knowledge-base/vector-embedding)).

A system running on an LLM OS is inherently less predictable than a conventional computer. However, it offers greater flexibility, enabling it to better adapt to new tasks and environments, potentially outperforming the capabilities of state-of-the-art systems by orders of magnitude.

Modern implementations of this idea are commonly described in terms of [AI agents](https://promptmetheus.com/resources/llm-knowledge-base/ai-agent), [tool calling](https://promptmetheus.com/resources/llm-knowledge-base/tool-calling-function-calling), [agent memory](https://promptmetheus.com/resources/llm-knowledge-base/agent-memory), and [agent orchestration](https://promptmetheus.com/resources/llm-knowledge-base/agent-orchestration). Protocols such as the [Model Context Protocol (MCP)](https://promptmetheus.com/resources/llm-knowledge-base/model-context-protocol-mcp) provide standardized interfaces for some of the tools and data sources envisioned by the LLM OS architecture.

## Sitemap

See the full [sitemap](https://promptmetheus.com/sitemap.md) for all pages.