Artifacts for every agent — open-source and self-hosted. Your agent builds a self-contained page, dashboard, or app and ships it to a live URL with one command — from Claude Code, Cursor, Codex, Cline, Aider, or any harness that runs a shell command. Then you review it in the browser and drop comments like a Google Doc, and the agent reads your comments and fixes it.
An open-source alternative to Claude Artifacts — except you host it, you own it, and any agent can drive it. No more screenshotting your agent's output and pasting it back into the chat.
agent builds → glance deploy → URL
↑ ↓
reads comments, fixes ← you comment in the browser
Self-hosted on Cloudflare's free tier — $0/month, you own the whole loop. Ships with a CLI and an agent skill, so any agent — Claude Code, Cursor, Codex, Cline, Aider — drives deploy → pull comments → reply → redeploy with no human in the copy-paste path.
Stack: Cloudflare Workers + Hono · React Router v7 · D1 · R2 · KV.
First enable R2 on your account (dashboard → R2 → accept terms — still free), then:
bun install
bunx wrangler login
scripts/setup.sh # provisions D1/KV/R2, deploys both workers, sets secrets, migrates, prints URL + tokensetup.sh is idempotent. At the end it prints a bootstrap token — open the printed /login, paste it into Complete setup, and you become the first superadmin. No Google account needed.
Multiple Cloudflare accounts?
export CLOUDFLARE_ACCOUNT_ID=<id>first. Manual provisioning, secrets, and optional Google SSO: see DEPLOY.md.
Pick a space, drop a folder, and your sites are live behind private/members/team visibility:
Superadmins get usage at a glance — users, sites, storage, page views, comments, and CLI activity.
Glance is also a home for audio — and the review loop works by voice.
- Serve & play — audio files (
mp3/wav/m4a/ogg/flac/aac/webm) serve with the right MIME type and HTTP Range, and render in a dedicated player (not the sandboxed HTML iframe), with page-anchored comments and a[m:ss]timestamp-insert shortcut. - Record → URL — tap the mic on the dashboard, record (live waveform, pause/resume), name it, and it deploys and opens straight in the player. Uploading a file is still one tap away.
- Voice comments — record a voice note right in the review composer (and in replies). It's stored, transcribed best-effort with Workers AI (Whisper), and shown as a voice card: inline player + transcript + badge. The transcript is the comment body, so the CLI/agent loop still reads everything as text.
Audio sites carry a mic badge across the dashboard, and glance comments prefixes voice comments with [voice] in the digest.
curl -fsSL https://glance.your-subdomain.workers.dev/api/install | sh # installs to ~/.local/bin/glance
glance login # device-code flow, opens browser
glance deploy <path> # file or folder → publishes to your personal spaceThe installer bakes in your instance URL and installs the agent skill so coding agents can drive the CLI. Any agent that can run a shell command drives Glance by calling the glance CLI directly — it's harness-agnostic.
The bundled skill teaches your agent to drive Glance. Install it into any harness — Claude Code, Cursor, Codex, OpenCode, Amp, and more — with the skills.sh installer:
npx skills add plivo-labs/glance # installs the glance-cli skill universally (Codex, Cursor, OpenCode, Claude Code …)The curl … /api/install | sh line above already installs the skill for Claude Code alongside the binary. The skill only wraps the glance CLI, so any shell-capable agent works with or without it.
| command | what it does |
|---|---|
login |
device-code flow, saves token to ~/.glance/config.json |
deploy <path> [--space <slug>] [--name <slug>] [--visibility <v>] |
uploads a file or folder (folders recurse, skip .git/node_modules) |
list |
your sites, with visibility + URL |
comments <space/slug> |
pull a site's review comments (voice comments show as [voice]) |
reply <thread> |
reply to a comment thread from the terminal |
delete <space/slug> |
confirms, then deletes |
move <space/slug> <new-space> |
moves a site (keeps files/comments/shares; URL changes) |
upgrade / version / logout |
self-update · print version · revoke session |
Defaults: --space = your personal space · --name = file/folder name slugified · --visibility = team (private · members also available). Point at another instance with GLANCE_API_URL=https://… glance <cmd>.
The CLI keeps itself current (once-a-day background check, atomic in-place swap). Opt out with GLANCE_NO_UPDATE=1.
- Uploaded HTML/JS is untrusted — served from a separate content origin (
CONTENT_URL), so app session cookies never reach it. This is why Glance stands up two Workers, not one. - Gated links carry short-lived, single-use HMAC tokens. Every tier requires an authenticated user — there is no public/anonymous access.
- Markdown renders with raw HTML neutralized under a strict CSP, so injected
<script>is inert.
Hosted sites can get browser-callable persistence — no keys, no config. Off by default; an operator enables it per deploy (see SHARED_BACKEND.md).
// Injected automatically when the site is opened through the Glance app; every request is
// brokered by the parent frame, so the page never holds a credential.
const notes = glance.db.collection('notes')
await notes.create({ text: 'hello' }) // create · list · get · put · deleteDocs are JSON ≤100KB in named collections. Every viewer can create and read their own docs; shared-* collections are readable by all viewers; the site owner reads everything and can moderate. Access is re-checked live, so revoking a share cuts data access immediately.
packages/api Hono Worker — /api/* + file serving, ships the React app as static assets
packages/web Vite + React Router v7
packages/cli `glance` CLI (Go)
Local dev: bun install && bun run db:migrate:local && bun run dev (main :8787 + content :8788 + vite :5173), then open http://localhost:5173. CI auto-deploys both workers on push to main.