Downloads

Four things are installable from this project: the agent skills, which are prompts an agent loads and needs no runtime; llmsx, a Python library and CLI for the llms.txt standard; llmsx-skills, the same reading surface for JavaScript; and the terminal browser over the concept tree, which today ships inside llmsx rather than as its own binary.

Agent skills

Installed with the skillsCLI, which reads them straight out of this repository — there is no tarball to download and nothing to build. Each skill is a single SKILL.md; installing one copies that file into your agent's skills directory.

# one skill, into the current project
npx skills add mithudso/llms-explorer --skill crawl-repo-to-llms

# globally, without prompts
npx skills add mithudso/llms-explorer --skill crawl-repo-to-llms -g -y

# see everything on offer first
npx skills add mithudso/llms-explorer --list

Eighteen are installable, seventeen of which have a page here. The six below are the ones worth reading first; the full catalogue lists the rest, and every page ends with the exact install line for that skill.

`crawl-repo-to-llms`	compile a whole repo into an agent dossier
`crawl-to-llms-txt`	condense a site or repo into an llms family
`llms-concept-abstractor`	pull one concept out of a corpus
`llms-deep-optimizer`	audit and fix an existing llms family
`concept-family-explorer`	map the family around a subject
`full-suite`	the whole pipeline, end to end

`llmsx` — Python library and CLI

A dependency-free, read-only CLI over the concept tree this site renders, plus a parser for the llms.txt standard v2. Python 3.11 or newer, MIT, zero runtime dependencies.

Not on PyPI yet — version 0.1.0 installs from source. The published pip install llmsx arrives with the metered CLI verbs at GA; until then this is the honest command:

git clone https://github.com/mithudso/llms-explorer
cd llms-explorer

# the library and CLI
python -m pip install -e llmsx

# with the terminal browser
python -m pip install -e 'llmsx[tui]'

llmsx --help

`llmsx-skills` — npm package

The JavaScript counterpart: ESM, Node 20 or newer, MIT, zero dependencies.Not on npm yet — version 0.1.0 is consumed from the repository, either as a path dependency or by importing it directly out of a clone.

git clone https://github.com/mithudso/llms-explorer

# from your own package.json
npm install ./llms-explorer/llmsx-js

# then
import { parse } from "llmsx-skills";

The terminal browser

There is a TUI today, and it is not a separate download: it ships insidellmsx as llmsx tui, behind the [tui] extra. It is the same Textual screen as the hub manager's Concepts tab — a tree, a filter that shows only matching branches, a detail pane, and frontier concepts drawn dim italic — reading the same generated tree this site serves at /tree/.

python -m pip install -e 'llmsx[tui]'
llmsx tui

What it deliberately does not do: the hub manager's two write actions —n to queue a concept and R to launch research — are absent, because they mutate a hub that a read-only package cannot reach.

Standalone, later

The planned standalone llms-explorer TUI is the same screen with its data source swapped: pointed at the API instead of a generated file, which is what makes the two write actions possible and turns the browser into something that can queue work rather than only read it. It is designed but not implemented — the hub manager's Concepts tab stays the reference implementation until it lands. No download exists yet, and this page will carry it when one does.

Two different "concepts"

llmsx tui and llmsx concepts tui browse two different data models, on purpose. The tree (llmsx tree / llmsx tui) is this site's own research concept tree — see this site's llms family for what it is built from. Concept packs (llmsx concepts) are<slug>.llms/ directories theconcept abstractor builds around one named concept: a manifest.json, a concept-graph.json of related terms, and the familiar llms.txt / llms-full.txt /llms-small.txt / llms-facts.txt /llms-vocabulary.txt ladder — pointed by default at~/.global-ai-hub/llms-concepts, overridable with --data or$LLMSX_CONCEPTS_PATH.

Running a skill from the CLI

llmsx family and llmsx optimize are thin wrappers overllmsx.skills.run_skill: one model turn holding the named skill's own instructions as the system prompt, your input as the user turn. Both print a one-line disclaimer before the model's output, because it is genuinely not the same thing as runningconcept-family-explorer orthe full suite end to end — those orchestrate multi-pass loops, subagent fan-out and concept-tree writes that a single API call does not drive. Useful for a quick read on a topic or a file; not a substitute for the loop when you need the loop.

Reading the site itself

The site is also a download, in the sense that matters to an agent: fivellms files that describe every page here, served astext/markdown and refreshed on every build.