Artifold
Your AI artifacts have a home now. Index, search, preview, share with one click, and use your past work as the style guide for your next one.
"I've started preferring HTML as an output format instead of Markdown. The added expressiveness means I get overall better output, and the chance of someone actually reading your spec, report or PR writeup is much higher if it's in HTML. … When writing this article, I asked Claude Code to read through my code folder and find all the HTML files I've generated, group and categorize them …"
— Thariq, The Unreasonable Effectiveness of HTML (Claude Code team)
Artifold is that library. If you're shifting from Markdown to HTML for specs, reports, designs, prototypes, throwaway editors — the way Thariq describes — Artifold is where they live.
You've been making a lot of HTML with AI lately: Claude Artifacts,
ChatGPT Canvas, v0, Lovable, Cursor. They land in ~/Downloads or some
project folder, you bookmark a tab, you mean to come back to that ROI
calculator you made three weeks ago — and you can't find it.
That's the whole reason this exists. I made a 30-day workout tracker for my partner, lost it in a maze of folders, regenerated a worse version, and decided to just build the index myself.
Local-first, fast, runs entirely on your machine. Plus the part where, when you do want to share an artifact, it's one click to a permanent public URL.
Three things you can do
1. ✨ Share any artifact in seconds → public URL anyone can open
Click any card → hit the share icon → in ~30 seconds you get back a
permanent URL like https://you.github.io/artifold-share/abc12345.html
that you can paste into iMessage, Slack, email, anywhere. URL is copied
to your clipboard automatically.
Free forever (uses your GitHub Pages quota — no Artifold infrastructure, no sign-up, no per-share fees, no expiry). Anyone with the link sees the fully-rendered report; recipient needs nothing installed.
2. 📚 Browse everything you've made, in one place
Sidebar filters by tool, status (shared / local), category, date. ⌘K palette to jump to anything by name, prompt, or intent. Click → in-app preview pane, no tab spam. Dark/light themes. Live auto-rescan — drop a new file in any watched folder and it appears within seconds.
3. 🎨 Close the loop with whatever design skill you use
Artifold indexes HTML. It does not care what wrote it — Claude Artifacts, ChatGPT Canvas, v0, Cursor, or any of the design skills people install on top of them. What changes between them is how much meaning survives.
| Skill | Artifold reads |
|---|---|
/craft (bundled, see below) |
everything: intent, conceit, four rotation axes |
| Hallmark | layout, theme and brief, from its CSS stamp + .hallmark/log.json |
| taste-skill, huashu-design, anything else | palette, fonts, tokens, skeleton, thumbnail, full-text search |
artifold skills # which of these you have, and what Artifold gets from each
The bottom row is not a consolation prize — it is most of the product. Fingerprinting is pure regex over the HTML and needs no cooperation at all.
The top row exists because a page can just say what it is, in four lines
of <head>. If you write a design skill, that convention is documented in
docs/ARTIFACT-METADATA.md and it is not
Artifold-specific. Skills that stamp their own format instead are read
through adapters in artifold/adapters.py; PRs for
new ones are welcome.
Worth knowing: of the three most-starred design skills, only Hallmark records anything between runs, and its log is per-repository. If your artifacts are scattered one-off files rather than one codebase, a machine-wide library is the only thing that can tell you what you already made.
The bundled /craft skill
→ See the side-by-side gallery:
four prompts run through claude-sonnet-4 twice — once plainly, once via
/craft. Same model, same prompt, visibly different output.
/craft is Artifold's reference implementation of a skill that instruments
its own output. It is optional — use Hallmark or taste-skill instead if you
prefer their look, and Artifold still works. Type /craft a 30-day strength tracker for a beginner in any session. The skill:
- Reads your library to see styles you've used, then deliberately picks a different direction so your next artifact doesn't look like the last one (the "all AI output looks the same" problem, actively fought)
- OR inherits a specific style:
/craft a poker probability explainer, like dobble— pulls dobble's actual CSS as the design baseline - Applies 12 opinionated design principles distilled from Refactoring UI, Linear, and Vercel/Geist (every one cited)
- Avoids 15 specific AI-slop signatures (purple-gradient hero, identical bento cards, decorative emoji on every list item, glassmorphism, etc.)
- Saves to
~/artifold-inbox/2026-05-26-<topic>.html— auto-indexed in Artifold within ~2 seconds
Your past work becomes your style guide for the next one. The loop closes.
Install
🪄 Easiest: ask Claude Code to do it
If you have Claude Code, paste this prompt — Claude figures out the rest, installs everything, and walks you through first-run:
[!TIP] Copy this into Claude Code:
Install Artifold from https://github.com/shubhamgoel27/artifold using pipx (or pip if pipx isn't installed). Then run `artifold init` and help me pick a folder to watch. After that, run `artifold install-skill` to set up the /craft skill. Open the dashboard when ready and tell me what to try first.
That's the whole install. No terminal commands to memorize, no dependencies to debug — Claude handles the awkward parts.
Or do it yourself
If you'd rather drive:
pipx install artifold # or: pip install artifold
artifold init # interactive wizard
artifold # serves dashboard + opens browser
artifold install-skill # adds /craft to ~/.claude/skills/
Don't have pipx? Run brew install pipx (mac) or python -m pip install --user pipx (anywhere), then the above.
After install, in Claude Code
/craft a one-pager comparing three SF apartments
/craft a 30-day strength tracker, like my last one
/craft a probability explainer for poker, in the style of dobble
(Restart Claude Code once after artifold install-skill so the skill loads.)
The first run installs Playwright's headless Chromium (~170 MB) for artifact thumbnails. After that, only new/changed files re-shoot.
What it actually does
Auto-indexes every *.html in your watched folders. Groups
-v2, (1), print variants into one card with a version dropdown.
Skips templates, .git repos, and anything buried 3 levels deep — your
library stays clean even when your folders aren't.
Source-aware — fingerprints Claude / ChatGPT / v0 / Lovable / Bolt
/ Gemini artifacts from HTML markers, tags each card with the tool that
made it. Reads <meta name="artifold:*"> tags from any skill that emits
them (the convention), and reads other skills'
own formats through adapters — Hallmark's CSS stamp
and .hallmark/log.json today.
Visual — every artifact gets a real screenshot thumbnail. Click a card → slide-out preview pane with tabs for Provenance (source URL, prompt, model, tags) and Design (palette swatches, fonts, mood flags).
Live — artifold serve watches your folders. Drop in a new artifact
and it appears in the dashboard within ~2 seconds, no refresh.
Searchable — ⌘K palette runs across titles, prompts, intents, and exposed actions (toggle theme, switch view, rescan, import). Linear / Raycast pattern.
All commands
artifold # default: scan + serve + open browser
artifold init # interactive setup wizard
artifold add <dir> # watch another folder
artifold roots # list watched folders
artifold scan # one-shot rescan
artifold open # open the existing dashboard
artifold share <file> # publish to public URL + copy to clipboard
artifold share --list # all shares
artifold share --revoke <id> # take a public share down
artifold import <url> # fetch a public Claude/v0/Lovable share URL
artifold link <file> --tool claude --source URL --prompt "..."
artifold info <file> # show provenance for a file
artifold designs # list design fingerprints
artifold designs --json --axes --limit 12 # recent rotation axes, slim (what /craft reads)
artifold designs <id> --template # dump CSS + skeleton (paste into Claude)
artifold inbox [topic] # print the canonical path for a new artifact
artifold skills # design skills, and what Artifold reads from each
artifold install-skill # install the bundled /craft into ~/.claude/skills/
artifold doctor # check setup; tells you exactly what to fix
Config
~/Library/Application Support/artifold/config.json on macOS,
~/.config/artifold/config.json on Linux:
{
"roots": ["/Users/me/Downloads", "/Users/me/work"],
"allow_repos": [], // dirs with their own .git to include anyway
"max_depth": 3,
"drop_dir": null, // where `artifold import` saves (default ~/artifold-inbox)
"enable_intent": false, // opt-in LLM intent metadata (Claude Haiku)
"categories": { // extend the auto-tag keywords
"Research": ["paper", "experiment", "ablation"]
}
}
Cache (thumbnails, manifest, dashboard, Playwright Chromium) lives
under ~/Library/Caches/artifold/. Wiping it just regenerates everything
from your real files — cache is replaceable, your source files are sacred.
Keyboard
| Key | What |
|---|---|
⌘K / Ctrl+K |
palette (actions + artifact search) |
↵ |
open selected in preview / run action |
⇧↵ |
open selected in new tab |
Esc |
close palette / preview |
/ |
focus the search box |
| Click card | open in preview pane (in-app, no tab spam) |
⌘/Ctrl-click |
open in a new browser tab |
Optional: AI intent layer
Artifold's core is fully local — no LLM, no network, no API key required.
If you want richer metadata (a one-line intent per artifact, topic tags, audience detection — used for smarter search):
pipx install 'artifold[intent]'
export ANTHROPIC_API_KEY=sk-ant-...
artifold scan --intent
~$0.003 per artifact with Claude Haiku, cached forever by content hash
so re-scans are free. ~$0.05 for 15 artifacts. Toggle off with
artifold scan --no-intent. Skip the extra and the feature simply
doesn't appear.
What Artifold is not
- Not a cloud product. Nothing leaves your machine unless you
explicitly
artifold share. There's no sign-up, no account, no Artifold server somewhere. Your library is~/artifold-inbox/and the dirs you pointed it at — that's it. - Not a replacement for git or your existing organization. It's a lens on whatever you already have.
- Not opinionated about where your files live. Multi-root by
design. Want it to watch
~/Downloads+~/Documents+~/work/reports? Runartifold addthree times. - Not a design skill, and not competing with one. That category is crowded and good — Hallmark, taste-skill and huashu-design are all worth installing. Artifold is the layer underneath: the place their output lands, gets indexed, and becomes findable a month later. Pick whichever skill you like the look of; the library is the same either way.
- Not trying to be everything for everyone. Built for the specific pain of "where did I put that thing I generated last month."
Why "Artifold"
A artifold is a working collection of pages that informs your next piece of work. Your past artifacts become the reference set for the next one. It's not an archive (cold storage), it's a working library.
The PyPI package is artifold because artifold was taken. CLI command is
artifold. Same pattern as open-interpreter / interpreter.
Status
v0.5.2, alpha — live on PyPI. Tested on macOS Sequoia; Linux should
work; Windows untested. Single-developer project made in evenings —
issues + PRs welcome, but I'm shipping what I personally use rather than
what's broadly polished. The /craft gallery
shows side-by-side outputs (default Claude vs /craft) on four prompts
through Opus 4.7 — if you want to see what the skill actually does
before installing, start there.
If you try it and something feels off, open an issue — even one line is helpful, "the X button is confusing" is exactly the feedback that improves things.
Roadmap
In rough priority order:
- Mobile dashboard — currently breaks below 760px
-
artifold adopt <file>— opt-in consolidation into~/artifold-inbox/for existing files (keeping multi-root for those who want it) - Cloudflare Pages backend for
artifold share(alternative to GH Pages for users withoutghCLI) -
artifold generate --like <id>— direct one-command generation, opt-in via[intent]extra - Markdown rendering — first-class support for
*.md - Semantic search — when you remember the gist but not the title
- Version diff view — when you iterate v1/v2/v3, see what changed
- A real test suite — currently human-tested
Made by
@shubhamgoel27 — built because I genuinely needed it. If you find it useful, starring the repo is the single best thing you can do so other people building with AI find it.
The /craft skill took real research to make non-generic — the 12
design principles trace to Refactoring UI chapters, Linear's
Method, Vercel's
Geist, and a couple of recent AI-slop
critique articles. If you ship cool reports with it, tag me — I love
seeing what people make.
License
MIT
1 maintainer
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file artifold-0.10.0.tar.gz.
File metadata
- Download URL: artifold-0.10.0.tar.gz
- Upload date:
- Size: 125.3 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via:
uv/0.5.14
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
93236a9a8e06831146b1af89c8321c3e1d6f0390aaa4b0630fce2fd4d1a20880
|
|
| MD5 |
220728b76e2faec850a9e6fc3f07aac6
|
|
| BLAKE2b-256 |
5846f38a71576bfc36dd30ac6a9107062fa736f14e7c79fe0a0f97519bcff1a8
|
File details
Details for the file artifold-0.10.0-py3-none-any.whl.
File metadata
- Download URL: artifold-0.10.0-py3-none-any.whl
- Upload date:
- Size: 141.1 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via:
uv/0.5.14
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
aa6388596355db35778f1f451bd93fafdf78427872392283de399ce4ff1927dc
|
|
| MD5 |
593df910bb361eec92913cc6f03cc705
|
|
| BLAKE2b-256 |
bc4a7890af7833701d33b7bee219772acc0f134a3986c14ba3ad9549195963e8
|
