pine

Pine 0.13.0 · Documentation

Hosting the Pine landing page

Back to Pine · Documentation

Build the site for its final public URL, then host the complete dist/site/ directory. Node's built-in modules perform the build; no package installation, Pine server, model credential or runtime service is needed. The build copies playback assets, renders the Markdown guides and sets canonical URLs, social metadata, alternate Markdown links, discovery files and host routing together.

Build once for the chosen address

From the standalone Pine repository, use Node 22.19 or newer:

node scripts/build-site.mjs --base-url https://johnlindquist.github.io/pine --lastmod 2026-10-25
node scripts/site-agent-check.mjs dist/site

Set --base-url to the final HTTPS address, including /pine for this GitHub Pages project. For a root-domain host, pass just that host's HTTPS origin. Use --lastmod for the actual date you changed the published content; the example date is this content's revision date, not a request to claim a fresh edit on every deployment. An optional --out chooses a different output folder. The build refuses to overwrite an unrelated directory.

The same source supports GitHub Pages, here.now and Vercel. Rebuild when the public address changes. Do not edit canonical links or social URLs separately. Upload the entire output, including .nojekyll, media/, vendor/, the guides, robots.txt, llms.txt, sitemap.xml and vercel.json where that host uses it.

Directory checking validates the files. It reports HTTP statuses and header negotiation as not measured; those require a running URL. The checker emits JSON, exits 0 when its applicable checks pass, exits 1 for failed checks, and exits 2 for an invalid invocation or an unreadable target. It is a local reproducible checker, not an Is Agentic score or a search-ranking test.

Review over real local HTTP

Serve the generated files with the delivered route interpreter:

node scripts/build-site.mjs --serve dist/site --profile vercel --port 8080

Keep that process running. In another shell, for the GitHub Pages base above:

node scripts/site-agent-check.mjs http://127.0.0.1:8080/pine/

The path prefix comes from the build's base URL. A root-domain build uses / instead. Open the same address in a browser. Inspect desktop and phone widths, follow Docs and Install Pine, read About, Contact and Privacy, and check a missing page. Play each workflow, seek while paused, switch tabs, and inspect the no-JavaScript view. A screenshot or directory pass does not establish that a deployed server routes requests correctly.

The vercel profile is a local emulation of the generated ordered routing rules. It uses a real HTTP server and the delivered vercel.json; it is not Vercel's CDN and does not verify a production deployment. Vercel's configuration documentation says header has conditions are not supported in vercel dev, so that command alone cannot prove Markdown negotiation.

To review the static-host limitation separately, stop the first process and run:

node scripts/build-site.mjs --serve dist/site --profile static --port 8080

In another shell:

node scripts/site-agent-check.mjs http://127.0.0.1:8080/pine/ --profile static

Static mode reports missing negotiation, Vary: Accept and a negotiated Markdown 404 as expected host limitations. Those checks are still visible in its JSON report. Do not describe that result as full negotiated-host acceptance.

Optional GitHub Pages

github-pages.yml is a dormant, opt-in example. Nothing under site/docs/ runs as a workflow. The repository owner can copy it to .github/workflows/site.yml, select GitHub Actions in Settings → Pages, review the github-pages environment and manually start the workflow. There is no automatic push trigger. The workflow asks for the public base URL and the content modification date, builds and checks the files, then uploads only dist/site/ with GitHub's Pages artifact actions.

The example targets the standalone Pine repository. In a combined workspace, set the shell steps' working directory to pine and the upload path to pine/dist/site. Keep the deployment environment and required pages: write and id-token: write permissions.

GitHub Pages can serve the HTML pages, explicit Markdown twins and custom 404.html. It cannot run this Vercel routing configuration or select a response from the Accept header. The page's alternate link gives agents a direct Markdown address, but that is a different capability from negotiation. Verify its actual Markdown content type and 404 status after publication.

A project site at /pine/ can publish /pine/robots.txt, /pine/llms.txt and /pine/sitemap.xml. Crawlers that request only origin-root discovery files will look at the account's root site instead. If the repository owner also controls that site, they may link the project sitemap there. A project-level robots file does not control the whole johnlindquist.github.io origin.

here.now

Publish the complete built directory using the manager's existing here.now workflow, with its final here.now URL supplied to --base-url. This is a static hosting path with the same negotiation limitation. Confirm the host preserves the explicit Markdown files and returns a real 404 for an unknown address; custom 404 handling depends on the host. Use the static profile when checking the published URL and read every reported limitation.

Vercel

Build for the intended stable Vercel public URL. Use the generated dist/site/ as the static deployment's project root, with its generated vercel.json at that root. There is no framework build inside this directory. The manager can publish that prebuilt directory through their normal Vercel workflow. A source repository deployment that ignores the output's configuration will not gain negotiation; verify the effective root before publishing.

The ordered routes match every page before the filesystem phase. An explicit Accept: text/markdown request with a nonzero quality value selects its Markdown twin. Ordinary HTML requests retain HTML. Both variants have Vary: Accept. After real assets have a chance to resolve, unknown Markdown requests route to 404.md with status 404; other unknown requests route to 404.html with status

  1. Explicit Markdown, XML and text files have their own content types. A

catch-all rewrite returning status 200 would not satisfy the error contract.

The public site lives at https://pine-pi.vercel.app. Its read-only Pine API and MCP server is added to the built directory before publishing. That step reads the release facts from the published GitHub release:

node scripts/build-site.mjs --base-url https://pine-pi.vercel.app --lastmod 2026-10-25 --out dist/vercel
node scripts/build-agent-api.mjs --site dist/vercel --base-url https://pine-pi.vercel.app --tag v0.7.0
vercel deploy dist/vercel --prod

Run the checker against the actual public URL after the manager publishes it:

node scripts/site-agent-check.mjs https://your-pine-site.example

Replace the example address with the real deployed URL. Inspect the returned page statuses, negotiated bodies, Content-Type, Vary, canonical URLs, sitemap and error responses. Run an Is Agentic scan only after that public replay. Search discoverability and scanner applicability remain external measurements. The read-only Pine API and MCP server run only on the Vercel deployment; optional exported team-agent applications are separate deployments with their own contracts.

Sources