Open-source starting points for business systems. Browse their tables, skills, agents and workflows, preview example records, and see what each package includes for a TulipFarm workspace.
The catalog includes marketing content, sales CRM, support desk, leave management, project tracking and product backlog packages. These are starter systems, not full replacements for Jira, Linear or Zendesk.
Catalog covers use purpose-specific illustrations, not execution diagrams.
The field previews come from the package definitions; full example records and
grouped resource lists remain on each package's detail page. The main action is
Install to workspace, which opens
https://app.tulipfarm.ai/_chat?intent=install-package&packageName=marketing-content&packageVersion=1.0.0
with the selected package's exact name and version. TulipFarm signs the person
in if needed, selects their default organisation and opens its private-chat
installer. Inputs and one whole-package approval are collected there; clicking
the marketplace link alone does not install anything.
JSON remains available as technical source, not the primary user journey.
The canonical origin is https://packages.tulipfarm.ai, configured in
src/lib/site.ts. Each package has its own search description, social image,
usage guide, limits, questions and related links in src/lib/package-content.ts.
The HTML includes canonical, Open Graph, Twitter card and JSON-LD metadata.
JSON-LD describes the catalog and package source, not an installed application
or a paid offer.
Local and default builds are noindex, with a disallowing robots.txt and an
empty sitemap. For a production deployment to the canonical domain, build with
SITE_INDEXING=true pnpm build. If CF_PAGES_BRANCH is present, it must be
main. Never use the indexing-enabled output on a preview host.
The production sitemap contains the homepage and all package detail pages,
not JSON endpoints or the 404 page. Deployment must serve the 404 page with
HTTP 404 and preserve the canonical trailing-slash routes.
Social PNGs are committed assets, so normal builds need no image tool.
After changing their copy, regenerate them with
node scripts/render-social-images.ts (requires ImageMagick's magick command).
The images are 1200 by 630 pixels. Search indexing and social previews still
need live verification after deployment; local metadata does not prove either.
Use Node 26 and pnpm 11.5.3:
pnpm install --frozen-lockfile
pnpm dev --host 127.0.0.1 --port 14321Open http://localhost:14321. This repository builds a static Astro website; it has no database, sign-in or installation backend.
pnpm validate:packages
pnpm check
pnpm lint
pnpm build
pnpm test
pnpm preview --host 127.0.0.1 --port 14321Cloudflare Pages project tulipfarm-packages is connected to this repository.
Pull requests receive noindex previews; merges to main build and publish to
https://packages.tulipfarm.ai. The older Pages project named packages is
unrelated and must not be used.
The Pages build uses Node 26.5.0, PNPM_VERSION=11.5.3 and
SKIP_DEPENDENCY_INSTALL=1, with this command and dist output directory:
pnpm install --frozen-lockfile && pnpm lint && pnpm check && pnpm build && pnpm testPages provisions the pinned pnpm version before this command; do not install pnpm globally over its Corepack shims.
Set SITE_INDEXING=true only in the production environment and false in
previews. GitHub CI also checks default, production and preview indexing.
Do not bypass its required check when merging. Pages runs its own quality
gates before publishing; a failed build leaves the last deployment serving.
To roll back, select a previous successful production deployment in the Pages
dashboard and choose Rollback to this deployment, then revert the source
change through a PR.
The pages.dev aliases retain canonical metadata and receive
X-Robots-Tag: noindex. Host-based redirects are not supported by the static
_redirects file. No paid service or runtime Worker is required.
Every package uses the public
system-package v1 schema.
Package files live at packages/<name>/<version>/package.json. Catalog metadata
is separate from the strict package format.
The site publishes /catalog.json and one exact JSON download per version at
/packages/<name>/<version>/package.json. Download URLs in the index are
relative to the website origin. The public feed is
https://packages.tulipfarm.ai/catalog.json; resolve its url and page values
against that origin. The feed and versioned JSON allow anonymous cross-origin
browser requests. Package inputs are resolved in a TulipFarm
workspace by its installer, not on this website.
The contents list groups definitions by resource type, not execution order. The installer must check permissions, collect declared inputs, obtain approvals and verify the result. A schema-valid package is not proof of a successful installation. Browsing or downloading never creates resources or connects to an external provider.
Add a new version directory rather than changing published version bytes. Use synthetic records, scoped permissions and explicit requirements. Never include credentials, private workspace IDs or real customer data. Author package content under MIT, or preserve compatible licensing and attribution.
Update the discovery metadata, run the commands above, and open a pull request. Package assets are data: the website never executes bundled scripts. Schema and catalog validation run before every static build.
MIT, covering this website and the package definitions.