A framework for organizing your interests in a harness for AI.

View on GitHub Explore the system
The idea

You are not one thing

You're made up of competing interests — work, family, side projects, the person you want to become — each with its own priorities.

Life, organized as entities

Most tools flatten all of your competing interests into a single to-do list and hope for the best.

LOS takes a different approach: each part of your life becomes an entity with its own identity, goals, rules, calendar, and tasks. Entities can have sub-entities, each with its own weights and priorities — a hierarchy that mirrors how you actually think.

Then it puts AI to work inside that hierarchy. A task can be delegated to the right entity, an event can land on the right calendar, and a model can act on behalf of the right interest — without your constant attention.

Why "Layered"?

Every entity is a layer. Your root entity is you. Below it sit the things you care about, and below those, the things they care about.

Weights and priorities flow up and down the tree. The result is a system that holds more nuance than any flat list ever could — and that AI can navigate on your behalf.

With the advent of AI, the way people live, work, and organize is changing. LOS maps your interests in a way that represents you, so AIs can act in your interest — and ultimately become more than the sum of our agendas.

What it does

Built for an AI-assisted life

The framework is model-agnostic: bring your own LLM, your own agent, your own hardware. Designed to be self-maintaining and light enough for a Raspberry Pi, a cloud instance, or anything in between.

Web interface

A secure SSL app with a calendar, to-do list, agenda tree, and file editor. Designed for a 4K monitor, adaptable to 1080p.

AI calling

A model-agnostic engine that runs prompts per entity, with per-entity configuration, memory, and tool access.

MCP action server

Exposes tools — calendar, to-do, cron, files, WhatsApp, memory, history — to any LLM or agent that speaks MCP.

WhatsApp interface

Talk to your system from a messaging app. Send tasks, ask questions, and get answers back in plain language.

Cron daemon

A lightweight scheduler that runs date, cron, and interval-based entries across every entity's agenda.

Memory server

Stores and recalls conversations and context so your AIs remember what matters.

Architecture

A system that lives under /los

LOS is designed to live on its own always-on Linux machine — bare metal, a VM, or a cloud instance. Hosting under /los is intentional: it can sit on its own partition, encrypted if you like, separate from the rest of the OS.

Top-level layout

/los
├── sys/ # the LOS software (this repository)
└── <username>/ # your agenda — one user per LOS install

The anatomy of an entity

<entity>/
├── identity # who or what this entity is
├── goals # what it's trying to accomplish
├── omni # always-on rules — a system prompt
├── entities # sub-entities with optional weights
├── prompt # the prompt the AI runs
└── data/
    ├── calendar/ # events by year/month-day
    ├── todo/ # to-do and done items
    ├── cron/ # scheduled entries
    ├── ref/ # reference files and notes
    └── project/ # per-task notes and attachments

This skeleton is copied for every new entity, so the structure is consistent all the way down the tree.

Services

The moving pieces

The AI manager can be configured per entity — so privacy-sensitive parts of your life never touch a model you don't trust, while the rest run on whatever suits them best.

Service Path What it does
Web app ssl_server/app.py The primary interface — calendar, to-do, agenda tree, file editor
AI caller aicall/aicall.py Runs prompts per entity; model-agnostic, with memory and tools
Action server aicall_mcp/action_server.py MCP tool server (calendar, to-do, cron, files, WhatsApp, memory)
WhatsApp server whatsapp_mcp/whatsapp_server.py Messaging interface to the system
Cron daemon loscron/loscron.py Scheduled entries across all entities
Memory server memory_mcp/memory_server.py Conversation and context recall
How to use it

See LOS-AI in action

A walkthrough of setting up your first entity, delegating tasks, and letting the system work for you.

Video coming soon

Paste an embed URL into the iframe above when you're ready.

Tip: YouTube embeds work with https://www.youtube.com/embed/VIDEO_ID; Vimeo uses https://player.vimeo.com/video/VIDEO_ID.

Theory and jargon

Think of a store-bought agenda: a calendar, a to-do list, maybe some scratch pages. A complete agenda also has a defined identity and defined goals. That's what the identity and goals files are for.

Every entity can have a set of rules that always apply — guidelines, a system prompt, anything that should shape every decision. That's omni.

In practice, nobody keeps their agenda perfectly up to date. That's fine — the system is built to run and sustain itself without your constant attention. Jot notes into a scratch file (the ref/ directory is for exactly that), and let the AI keep the rest honest.

The bigger picture

LOS is a framework — agnostic and ambivalent to the model or agent. Today it manages your agenda. Tomorrow it can be the framework your AIs use to interact with each other, with cryptocurrency smart contracts, and with corporate and legal entities.

Your own AI lawyers working on your behalf, without your interaction.

Say goodbye to things falling between the cracks.

Ready to organize your interests?

LOS is in active development. Contributions are welcome. Clone the repository, create your user with /los/sys/bin/los-user-manage, and start the services.

Get LOS-AI on GitHub