Lolly 1.0.1

Lolly 1.0.1

First public open-source release — 2026-08-30. Free software under the Mozilla Public
License 2.0; the content below is brand-agnostic and reflects the current platform.

Lolly - public launch

Lolly is a constraint-first, template-driven platform that produces creative assets at studio quality and at scale - image, motion, audio, document and more - from simple inputs.
One platform-agnostic engine runs the same render path across a web PWA, desktop and mobile apps,
and CLI + TUI. Tools are data, not bundled code - a manifest, a template and optional hooks -
so new tools can ship without an app update. Lolly runs fully standalone on your own device.
Lolly is completely free and open source. It is licensed under the Mozilla Public License 2.0.

What makes this release

• 🥋 On-device by design. Rendering happens on your device. The on-device utilities (strip
  hidden metadata, compress a PDF, format/redact text) never upload your file - bytes in,
  bytes out, locally. Nothing phones home.
• 🔐 Content Credentials (C2PA) built in. Exports can carry tamper-evident provenance, and
  federated source assets keep their credentials through a render. Verify any file's
  credentials in-app or from an agent.
• ⚡️ One render path, everywhere. The web app, the CLI and the desktop/mobile apps share the
  engine, so a render is identical across them. Every input is expressible in the URL, and the
  CLI is that same URL under a different transport - GUI and automation never drift.
• 📐 Physical units and print output. width/height accept mm/cm/in/pt with a
  dpi control; PDF exports fully color managed with separation plates if needed.
• 🏭 Batch mode & Multi-edit. A spreadsheet-style grid generates many assets at once.
  A live preview of multiple assets editable at once, perform global or isolated edits visually.
• 🪆 Tool composition. A tool can live-render another tool as an asset, and tools are
  addressable as portable embed URLs.
• 👁️ Accessibility. Tune it quiet or loud - reduce-motion, high-contrast, large-text and calm-previews, or richer audio and visuals - opt in without changing export quality.
• 🧠 Governed automation via MCP. An optional Model Context Protocol server exposes the
  catalog and render path to AI agents (list, describe, build a link, render, transform,
  redact, verify) under the same rules as the app.

The tool set 🧰

The public catalog is brand-agnostic; a deployment mounts its own brand pack (or starts from a
neutral one) and every tool conforms to that brand's tokens automatically. Tools span:

• Everyday: QR codes (link, contact, Wi-Fi, event, location, text), quote cards, code-to-
  image, a recomposing dynamic layout, a global meeting planner.
• Design: an open canvas, charts (bar/line/area and D3-powered), a photo darkroom
  (halftone, scanline, posterize, duotone, dither, ASCII), mesh gradients, street maps.
• Utilities (on-device): strip hidden metadata from images and PDFs, compress a PDF,
  format/decode/hash/de-identify text, grow a colour palette with perceptual ramps and
  WCAG/APCA readability, capture a web page as a high quality vector image.

Tools are marked Official (brand-approved, watermark-free) or Experimental (exports
carry a PREVIEW watermark for testing until graduated).

Privacy and provenance

• Creations and files stay on your device by default; saved sessions and the asset cache live
  in local storage, and you can clear them per category.
• Exports embed authorship/provenance metadata per format, with no personal data unless you
  opt in.
• The on-device transform path never watermarks and never embeds provenance - a file you
  brought is yours.
• Powerful provenance detection including a genAI content assessment for text and c2pa marked files.

Security and licensing

• Licensed under MPL-2.0. Third-party notices are reproduced in
  THIRD-PARTY-NOTICES.md, with a machine-readable CycloneDX SBOM
  (CI-checked for drift).
• Security posture and disclosure process are in SECURITY.md. Tool hooks run
  with the host bridge injected but are not a security sandbox - run only tools you have
  reviewed until Worker isolation ships.

Get Lolly

• Run it in the browser - directly via the web or PWA.
• Download the app - for this release, from lolli.li (verify any file
  against SHA256SUMS.txt):
  • macOS (Apple silicon): signed + notarized .dmg.
  • Linux: .rpm (openSUSE / Fedora family),
    .flatpak (any distro),
    .deb (Debian / Ubuntu), and an Arch pacman
    channel - add [lolly] with Server = https://lolli.li/arch/$arch
    (SigLevel = Optional TrustAll) to /etc/pacman.conf, then pacman -Syu lolly-desktop-bin.
  • Android: .apk (sideload; no store).
  • iOS is pending App Store review; a sideloadable build exists on request.
• Build from source / self-host - clone, npm install, npm run dev:web (Node >=22.18 or

  =24); a first render takes about 60 seconds (docs/make-something.md).
  Self-host the built web shell for a team; full operator documentation lives under docs/.

• Governed at org scale (optional): pair a deployment with the open-source lolly.work
  control plane for SSO, feature-flag / export / watermark policy, catalog federation,
  approvals and a hash-chained audit log - served to the shell without a code change. Lolly
  still renders on-device: OSS = individual freedom, OSS + control plane = organizational
  freedom.

Known limitations

• Saved state is per-device; clearing browser storage loses saved sessions.
• Web-page capture is native on the desktop apps and via the companion Chrome extension on the
  web; the bare web shell and the CLI stub it until they are deployed with access to chromium.
• Hosted rendering of hook-heavy / HTML-heavy tools needs the optional Chromium worker tier;
  the fast path refuses them by default.