Roadmap and non-goals

What v1 does not do

Each of these is a decision with a reason, stated so you can tell whether it matters to you before you install.

Multi-locale routing

Full internationalisation means locale-prefixed routes, alternate language links in both the metadata and the sitemap, per-locale feeds, translation linking, and locale-filtered prerendering. That is a large surface, and it multiplies the number of pages the blog builds.

What was taken is the cheap half. Retrofitting locale into the content source interface later would be a breaking change to the one interface that exists to stay stable, so four optional locale fields are reserved on the schema and the contract now, unused. Adding locale later is additive rather than breaking.

Tailwind v3

Supporting it would mean a second colour format, a second prose bridge, and a second config shape. Decisively, v3 users have to pin an old shadcn CLI that predates namespaced registries, so half the install story does not function for them. Supporting v3 means shipping a second product against a CLI that cannot install it.

agentblog doctor reports this as an error rather than a warning.

An llms.txt in the blog

See why the blog does not ship one. Short version: Google is on record that Search does not use it, no major provider consumes it in production, and the best-instrumented measurement anyone here could retrieve logged 84 requests to /llms.txt out of 62,100 AI bot visits over 90 days.

This documentation site serves one, because developer documentation with programmatic readers is the one place the token saving is real.

Certified monorepo support

See installing in a monorepo, which states precisely what works, what is untested, and where each file lands. The mechanism works because shadcn already handles it. A fixture is not in CI yet.

Also not in v1

• A hosted CMS. The content source interface is the seam. There is nothing behind it yet.
• An analytics integration. The referrer classifier ships as a pure function with no dependency, and measuring AI traffic shows the wiring for three tools. Shipping an integration would mean picking a vendor for you.
• Comments. Not a blog problem worth solving twice.
• A theme gallery. The blog has no theme. That is the point.
• Search. Your site probably already has one, and if it does not, that is a site decision rather than a blog decision.

Order of work

Now. The registry, the blog, the CLI, and the agent layer. The MDX source is the only content source.

Next. Supabase and Convex sources, in that order, each passing the same eight contract assertions. A monorepo fixture joins CI alongside the first of them.

After that. Measurement. agentblog audit --crawlers reads a log you already have and reports verified hits per bot per week. A hosted version reads Search Console over OAuth. Both sit on one interface, so the free tier and the paid one are the same shape.

Later. A hosted content source, and a dashboard on top of the measurement layer.

Maintenance, as a standing job

This product depends on third-party surfaces that move. Naming the cadence is more useful than promising to keep up.

The fixture in CI tracks the latest stable Next.js, which means upstream breakage surfaces in the AgentBlog repository before it surfaces in yours.

Version compatibility

Next.js 17 will invalidate parts of this, the same way 16 invalidated the 15-era guidance. When it lands, the fixture moves first and the blog follows.