CLI_CODE

Discover the right CLI tool or AI model in seconds.

500 CLI Tools · 50 AI Models · 20 AI Projects

Search → Compare → Build Stacks → Get Commands → Run Models Locally

No backend · No account · No tracking · Open source

▶ Open the live demo

---

Search, compare and discover 500 command-line tools, 50 AI models and curated developer stacks. CLI_CODE is an open-source, client-side directory: everything runs in your browser from static JSON, so there is no backend, no account and no tracking.

Live demo

https://mughil.github.io/CLI_CODE/

Try this	Link
Browse all 500 CLI tools	registry.html
Describe a task, get tools	find.html
Compare ripgrep vs fd	compare.html?slugs=ripgrep,fd
Browse 50 AI models	models.html
Models you can run locally	run-local.html
Open-source AI projects	ai-explorer.html

Project statistics

Dataset	Count	Notes
CLI tools	500	500 unique ids, 500 unique slugs, 0 duplicates — enforced in CI
AI models	50	18 providers; 22 open source, 11 open weight, 13 API-only
AI GitHub projects	20	star counts fetched from the GitHub API and timestamped
Curated stacks	7	preset tool sets + capability matrices
Categories	53	derived from the data, not hand-maintained
Runtime dependencies	0	vanilla HTML/CSS/JS; Node is build + CI only

Key capabilities

• Weighted search across name, tags, summary and use cases — instant, client-side.
• Find my CLI — describe a task in plain language, get ranked tools and the reason each matched.
• Find my model — tick requirements (local, coding, open source, large context…), get deterministic scoring with a per-result explanation.
• Compare — 2–4 CLI tools side by side with the differences called out.
• Model Battle — 2–4 AI models compared on openness, license, context, modalities, pricing.
• Stacks — curated tool sets per discipline plus capability matrices.
• Cheat sheets — printable command reference for a set, stack or your favourites.
• Run locally — which models run on your own hardware, grouped by runner (Ollama, llama.cpp, MLX, vLLM…).
• Saved — favourites, compare queue and recently viewed, stored only in your browser.
• Keyboard-first — / to search, ↑↓ to move, Enter to open, Esc to clear.

Screenshots

Not checked in yet — see the live demo. Contributions welcome: place images in docs/screenshots/ and link them here.

Quick navigation

Live demo · Why it's different · CLI features · AI model features · Run models locally · Development · Data architecture · Contributing · Validation · Deployment · Changelog · Roadmap

Why CLI_CODE is different

Task-first	Find my CLI and Find my model take a goal or a checklist and rank by what actually matches — and show you why. Deterministic scoring, never disguised as an LLM.
Honest data	Derive only what is safe to derive; leave the rest blank. Verified facts live in human-maintained overlays with sources[]. Nothing is fabricated.
Open ≠ downloadable	AI models are labelled open source / open weight / proprietary by their actual licence, not by whether the weights can be pulled.
Real comparison	Compare and Model Battle lead with the differences that matter, not a metadata dump.
Private by construction	No backend, no analytics, no cookies. Favourites and history never leave localStorage.
Portable	Every path is relative — the site works at / or at /CLI_CODE/ with no configuration.

CLI features

Page	Purpose
index.html	Search-first landing page
registry.html	Browse — weighted search, 6 facets, sortable, keyboard-driven, URL-persisted
cli.html?slug=	Tool profile — install methods, platforms, use cases, alternatives, JSON-LD
find.html	Find my CLI — natural-language recommendations with explanations
stacks.html	Stacks — 7 curated tool sets + capability matrices
compare.html?slugs=a,b	Compare 2–4 tools
cheatsheet.html	Printable command reference
saved.html	Favourites, compare queue, recently viewed (browser-local only)

AI model features

Page	Purpose
models.html	Models — search + facets (provider, openness, availability, capability, context, local)
model.html?slug=	Model profile — capabilities, licence, pricing, alternatives, Sources, JSON-LD
find-model.html	Find my model — deterministic requirement scoring
model-compare.html?slugs=a,b	Model Battle — 2–4 models, key differences
run-local.html	Local-capable models grouped by runner
ai-explorer.html	20 open-source AI GitHub projects, filterable by category

Every model fact is traceable through sources[], and volatile values carry their own timestamps (pricingVerifiedAt, starsVerifiedAt, lastVerified).

Run models locally

34 of the 50 models can run on your own hardware. run-local.html groups them by runner — Ollama, llama.cpp, MLX, vLLM, Transformers, LocalAI, LM Studio, TGI — using only what the model card documents. Hardware requirements are never invented; where they vary by quantisation the page says so.

Development

fetch() needs HTTP — do not open index.html from the filesystem.

npm ci                     # clean, lockfile-exact install
npm run build              # regenerate all derived data + SEO blocks
npm run check              # schema + cross-refs + drift + links + HTML + models
npx http-server -p 4173    # or: python -m http.server 4173

Then open http://localhost:4173.

Optional, network-dependent (not run in CI):

node scripts/check-links.mjs --net        # reachability of every external CLI URL
node scripts/check-model-links.mjs --net  # reachability of every model/project URL
npm run model-stats                       # human-readable AI dataset roll-up

Data architecture

Two independent datasets, each with its own ids, slugs, schema, validators and statistics.

CLI tools

data/registry.json           upstream harness CLIs (79)        ─┐
data/public_registry.json    upstream public CLIs (24)          │
data/catalog_registry.json   hand-curated CLI catalog (397)     ├─ build inputs
data/registry-dates.json     last-verified dates               ─┘
data/overlay.json            human-verified facts, by slug  ── contributor-editable
        │  scripts/build-data.mjs   (deterministic, no network)
        ▼
data/clis.json               normalized entries the site loads      (generated)
data/meta.json               facet counts                           (generated)

AI models

data/models/*.json           per-group model records, each with sources[]  ── contributor-editable
data/ai-projects.json        GitHub AI projects (GitHub-API verified)      ── contributor-editable
        │  scripts/build-data.mjs
        ▼
data/models.json             merged, normalized                            (generated)
data/model-meta.json         facet counts + project meta                   (generated)

Never hand-edit generated files — data/clis.json, data/meta.json, data/models.json, data/model-meta.json, sitemap.xml, robots.txt and the <!-- SEO:auto --> blocks in the HTML are all rewritten by npm run build, and a drift gate in npm run check fails if they are out of date.

Validation

npm run check runs, in order:

Step	What it enforces
validate	JSON Schema 2020-12 for entries and stacks; unique ids/slugs; resolving alternatives/related; size gate: exactly 500 tools, 500 unique ids, 500 unique slugs, 0 duplicates
check:drift	Rebuilds in memory and fails if any generated file is stale
check:links	Missing/case-mismatched local refs, broken anchors, duplicate ids, missing alt, target="_blank" without rel="noopener", stray localhost/TODO/console.log
check:html	html-validate across every page
check:models	Model + project schemas, unique ids/slugs, openness enum, resolving cross-references, URL well-formedness

Expected output:

CLI DATA VALIDATION
Total tools: 500
Unique IDs: 500
Unique slugs: 500
Duplicate IDs: 0
Duplicate slugs: 0
PASS

Deployment

.github/workflows/deploy-pages.yml builds, validates and publishes to GitHub Pages on every push to main; a failing schema, dangling reference or drift blocks the deploy. One-time repository setup: Settings → Pages → Source: GitHub Actions.

Static assets are content-hash stamped (?v=<sha1>) at build time, so a returning visitor never runs a stale bundle against fresh data.

Contributing

Good first contributions: verify a tool's facts, add install examples, propose a stack, improve model sourcing. See CONTRIBUTING.md for the exact steps for:

• adding a CLI tool · correcting CLI facts · adding an AI model
• adding an AI GitHub project · editing a stack · validating · opening a PR

Also: Code of Conduct · Security policy · Model data rules · Roadmap

License

Apache-2.0 — see LICENSE.

The registry.json / public_registry.json snapshots come from the upstream CLI-Anything project. The curated catalog in data/catalog_registry.json is maintained in this repository — one real tool per entry, with facets derived by the build.

---

Found this useful? ⭐ Star the repo, open an issue with a tool or model we're missing, or send a PR.

Live demo · Issues · Contributing