llmos

[!IMPORTANT] Archived 2026-07-24. Dismantled by its author in commit 323e717 (2026-03-14), which moved robotics to skillos_robot and the dream engine to evolving-memory. The markdown-agent thesis continues in skillos.

**The docs/ book is stale** — it documents modules deleted in that commit and cites a test count (346) that no longer matches the 46 tests remaining. Not related to llm_os, despite the name.

LLMos

The Agent Kernel — where LLMs are the native runtime.

LLMos is an operating system for AI agents. It treats language models the way Unix treats C programs — as the native executable format. Agents are markdown files. The kernel is markdown. Skills, memory, and configuration are all markdown.

LLMos is the prefrontal cortex of a three-part cognitive ecosystem:

How It Works

The kernel runs entirely in the browser. Agents are markdown files with YAML frontmatter — the development LLM creates, reads, modifies, and orchestrates them. When the system learns a new pattern, it writes it into a skill file. The agent definition is the documentation is the evolution history.

Quick Start

Open ` The workspace launches with a chat panel, sidebar, and context viewer — everything runs client-side.

For the Electron desktop app:

Core Concepts

The Volume System

Skills flow upward as they prove reliable:

The Kernel

The ClientKernel orchestrates all components: