Lloyaldocs
lloyal.ai GitHub
Start

Quickstart

Three commands to an app with the model inside it. No API key, no Docker, no vector database.

On this page

You need Node.js 24 or newer and about 10 GB of memory. If you are not sure, System requirements has the checks and a three-step Node.js install.

Create and run

Terminal
npx lloyal-ai new my-app --template research
cd my-app
npm run dev:desktop

The first command creates my-app/, installs its dependencies and prints how to run each surface. Leave out the name and the flags and it asks for them instead: name, surfaces, model, template. (Run from a script or CI, where there is no terminal, it skips the install — run npm install in the folder.)

The first launch downloads and verifies the models the template names; research needs three — a 4B reasoning model, a reranker and a vision projector. The app shows each download with its bytes, rate and time left. Every launch after that opens at once.

Then ask it something worth investigating.

Pick a template

Template Flag What you get
Wiki --template basic (the default) A small Wikipedia research app: two agents and an article view. The smallest program to read and change.
Deep research --template research A grounded multi-agent investigation: plan, parallel inquiries, a cited brief, a library that later briefs build on.

Each generated project carries its own README with recipes for that template.

Pick a surface

The same program runs in all three. Run whichever you like from the project folder:

Surface Run The model runs in
A desktop app npm run dev:desktop an engine process the window talks to
A browser npm run dev:web a host you serve; the browser connects to it
Your terminal npm start the process itself

Working with a coding agent

Every project carries an AGENTS.md (and a CLAUDE.md pointing at it): the rules a coding agent needs before it edits the code — the ownership model, where each thing already lives, and what proves a change. Point your agent at it first.

Next

↑
Search