LifeOS

LifeOS is a layer that sits on top of an AI coding harness and gives it your context: your goals, the people in your life, your preferences, and your definition of done. The harness supplies the raw capability; LifeOS adds persistent memory across sessions, a library of skills, routing that triggers the right workflow from a plain request such as "research this", and self-modification based on what the system learns. The organising idea is moving from a Current State to an Ideal State, and the project describes its approach as intent engineering: capture what you are trying to achieve, convey it to the AI on every task, then verify the result against it.

Installation is performed by the AI itself. You paste a single prompt into Claude Code, Cursor, Codex, or Hermes, and the agent reads the setup page and walks through setup, asking permission before it touches anything. A one-line terminal installer exists for Claude Code on macOS and Linux. Either path needs a capable coding harness and bun.

The code is TypeScript and Bash. LifeOS is built on primitives that most harnesses share, such as hooks, skills, context files, and agentic routing, rather than on one vendor's features, so it is harness-agnostic by design; Claude Code is the most-tested path. It ships as one self-contained skill that bundles the whole library, including research, security, writing, and art skills. Named components include TELOS, the Algorithm, Arbol, Bunker, the ISA System, Cortex, Synapse, Atlas, Ledger, the skill and hook systems, Pulse, Voice, Learning, Security, and a Hermes sidecar.

LifeOS is positioned as complementary to Fabric: Fabric is a collection of prompt patterns for specific tasks, whereas LifeOS is the infrastructure for how a digital assistant operates, and Fabric patterns can be integrated into LifeOS skills. Upgrades leave the USER/ directory untouched and merge settings rather than overwriting them, so hooks, statusline, and custom configuration are preserved; the installer detects an existing setup and merges into it.

The roadmap lists local model support (Ollama, llama.cpp), routing different tasks to different models by complexity, remote access from mobile and web, outbound phone calling, and notifications over email, Discord, Telegram, and Slack.

Features

• Persistent memory: the assistant remembers past sessions, decisions, and learnings
• Personal context: goals, contacts, and preferences are available without re-explaining them
• Skill library: one bundled skill installs a library covering research, security, writing, art, and more
• Intent routing: a plain request such as "research this" triggers the matching workflow
• Self-improvement: the system modifies itself based on what it learns
• AI-driven setup: a single prompt makes the coding harness perform the setup, with a curl shortcut for Claude Code
• Harness-agnostic: built on hooks, skills, context files, and agentic routing rather than one vendor's features
• Safe upgrades: USER/ customisations are never touched, settings merge, and the installer detects and merges into existing installs