# On Prem AI Guide

> An independent guide to on premise AI: the four layers a real install buys — inference, agent runtime, data and memory, policy and audit — and the open-source runtime Kortix ships.

Every page has a markdown twin (linked below) and the whole site is in [llms-full.txt](https://onpremaiguide.com/llms-full.txt).

## Pages

- [Home](https://onpremaiguide.com/index.md): Inference, agent runtime, data and memory, policy and audit: what fills each layer of on premise AI, and the open-source runtime Kortix ships.
- [What it is](https://onpremaiguide.com/what-is-on-prem-ai.md): On premise AI means the model and the software around it run on servers you own, across four layers that each have an open-source option.
- [Server requirements](https://onpremaiguide.com/on-prem-ai-server-requirements.md): GPU memory and storage size the box; the open-source runtime, identity, approvals and audit are the rows your hardware quote leaves out.
- [LLM vs API](https://onpremaiguide.com/on-prem-llm-vs-api.md): On prem LLM vs API: the open-source trade swaps a per-token bill for a per-change upgrade and a capacity plan only your team can size.
- [Self-hosting](https://onpremaiguide.com/self-hosted-ai-agent-platform.md): A self-hosted open-source agent stack is only as local as the machine that runs the work, and five questions find out where that machine sits.
- [Security and audit](https://onpremaiguide.com/on-prem-ai-security-and-audit.md): An open source on-prem stack turns AI security into a permission model you can read: a record per action, a gate you set, and secrets off the session.

## Kortix

- [Kortix](https://kortix.com): The open-source AI Operating System.
- [Kortix docs](https://kortix.com/docs): installing, self-hosting and configuring Kortix.
- [Kortix on GitHub](https://github.com/kortix-ai/suna): the open-source repository.