ELI

ELI exists because the people who most need an AI assistant are exactly the ones cloud AI serves worst.

• People who can't justify another monthly subscription

• People who live with bad internet or none

• People who don't want their conversations sitting on a stranger's machine

• People who work with sensitive material they'd never paste into someone else's website

• People who just want to own their tools

The Longer Version

I Came at This from Physics and Engineering

I didn't come at this from computer science. I came at this from physics and engineering. I learned most of it the hard way, in the wrong order, on my own. A year of working on it nearly every day.

I built ELI because I looked at the AI landscape and saw a problem:

Every "AI assistant" is a tenant arrangement. Your data lives on their machines, under their terms, subject to their pricing, their outages, their policy changes, and their ability to read it.

ChatGPT, Claude, Gemini, Copilot — they're all thin apps talking to someone else's datacenter. You don't own the assistant. You rent it. And if the company disappears, changes the price, or reads your conversations, you have no recourse.

I wanted to build something different. Something that actually belongs to you.

The Technical Reason

I Wanted an AI That Never Fakes an Action

Most AI assistants are fundamentally dishonest. They "say" they did something, but they actually just generated text that sounds plausible. If you ask a cloud assistant to open an app or send an email, it might say "Done!" but it never actually did anything.

ELI is different. If ELI says it did something, it really did—and if it couldn't, it says so plainly instead of bluffing. This is built into the software, not just good manners. It's the "No-Fake-Actions" guarantee, and it's enforced in code.

I Wanted an AI That Actually Runs Locally

Most "local" AI assistants still phone home. They claim to be private, but somewhere in the fine print, they're sending your data to a server.

ELI is offline by default, enforced at the socket. When networking is off, a process-wide guard fail-closes at the socket. Even if some component tried to phone home, it physically can't.

I Wanted an AI That Can Actually Do Things

Most AIs are just text boxes. You type, they reply. That's it.

ELI has a body, not just a mouth. It opens apps, plays music, controls windows, takes screenshots, sees images and your live screen, reads text out of pictures, and can even click where your eyes are looking via webcam gaze tracking.

I Wanted an AI That Improves Itself

Most AIs are static. They never get better.

ELI logs its own failures. It can write, syntax-check, import-test, apply, and automatically roll back patches to its own code. It can fine-tune its own model on your conversations via LoRA.

The Philosophical Reason

ELI Is Ownership, Not Tenancy

"It runs with the internet cable pulled out. It keeps working if the company that inspired it vanishes. It costs nothing to run beyond your own electricity. It answers only to the person at the keyboard."

ELI is a statement about privacy, sovereignty, and self-determination.

ELI Is for People Who Can't Trust Cloud AI

For anyone who works on sensitive material—research, inventions, personal life, anything you would not paste into a stranger's website—cloud AI is not an option. It's a non-starter.

ELI is for those people.

ELI Is a Challenge to the Cloud AI Model

Most companies want you to believe that AI must run in the cloud. That it's too big, too complex, too expensive to run locally.

ELI is proof that they're wrong. 152,000 lines of Python. 209 capabilities. 15 agents. 12-stage pipeline. 4-layer memory. Self-healing. Vision. Voice. Gaze. All running on hardware you already own.

-- Jason (ShadowESC95)