LifeOS: Modeling Life as a Hill-Climbing Optimization Problem

Forum topic · ✨步子哥 · 2026-08-10

Summary

LifeOS, a trending GitHub project by Daniel Miessler, is a harness-agnostic general-purpose AI framework that uses hill-climbing optimization as a metaphor for personal development. The system formalizes the gap between Current State and Ideal State, with Euphoric Surprise acting as an internal reward signal. TELOS is a Markdown-based context file describing identity, goals, and history, giving the AI persistent memory of the user. Skills are reusable capability packages that codify recurring workflows. Installation runs through a single prompt or curl script, with the AI harness handling setup. The framework is deliberately portable across Claude Code, Cursor, Codex, and other harnesses, prioritizing longevity over feature lock-in. The article argues the real contribution is making a vague life objective function explicit, so AI can help users see terrain and avoid local optima rather than make decisions for them.

Overview

LifeOS is Daniel Miessler's new project, positioned as a "General Purpose AI Harness for Life and Work." It re-frames personal development as a hill-climbing optimization problem, using AI to help users move from Current State to Ideal State while tracking Euphoric Surprise as a reward signal.

Key points

Current State → Ideal State framework

Current State: Where you are now (skills, relationships, career, health, finances)
Ideal State: Where you want to be (same dimensions)
Euphoric Surprise: Operationally defined as the gap between expected and actual experience when moving toward Ideal State; serves as an internal reward signal

Hill-climbing metaphor and its limits

Hill-climbing is the oldest optimization algorithm: pick a neighbor higher than the current point, move there, repeat
Known flaw: traps at local optima because the algorithm cannot temporarily descend
LifeOS does not solve this problem; instead, it uses AI to surface nearby high-value states the user has not noticed

TELOS: a user-owned context file

Records identity, values, long-term goals, concerns, current location, destination, and lessons learned
Implemented as portable Markdown files rather than a database or API
Enables user-owned, version-controlled, git-diffable context that the AI can read on every interaction

Skills: reusable capability packages

Skills bundle workflows such as research, writing, security audits, and creative work
Convert recurring tasks into reusable, composable, iterable units
Distinguished from cangjie-skill (external knowledge distillation) or DevGraph (knowledge graphs) by focusing on personal workflows

Harness-agnostic architecture

Runs on Claude Code, Cursor, Codex, or any capable AI harness
Core primitives (hooks, skills, context files, agentic routing) are tool-agnostic
Designed as a portable layer rather than a vendor-locked application

Installation

Single prompt: Read https://ourlifeos.ai/install and install LifeOS for me.
Or one-line curl: curl -fsSL https://ourlifeos.ai/install.sh | bash

Design philosophy

The real difficulty in optimization is defining the objective function, not finding the optimum
LifeOS externalizes a vague personal objective function via TELOS, Current State, and Euphoric Surprise
AI's role is to widen the user's view of the terrain, not to climb on their behalf

References

Project: https://github.com/danielmiessler/LifeOS
Official site: https://ourlifeos.ai
Author blog: https://danielmiessler.com/blog/personal-ai-infrastructure