Compare
Rebilder and @vercel/agent-readability
Vercel’s package gives developers a free site audit and middleware that routes AI agents to markdown. Rebilder builds that clean version from your business data, then shows which agents visit and helps you answer what customers ask. Run the audit in CI and use Rebilder to serve, measure and improve.
Vercel’s open-source package for agent detection, markdown routing and a site audit. All comparisons
What each does
| Dimension | @vercel/agent-readability | Rebilder |
|---|---|---|
| What it is | An MIT-licensed npm package with framework middleware and an audit CLI. The latest version on September 28, 2026 was 0.7.0. | A free, self-hostable gateway (Apache-2.0), a public scanner and a Console for tracking, answers and Autopilot. |
| Which spec it checks | Vercel’s Agent Readability Spec, a guide on vercel.com now titled “Make your site readable by AI agents”. | The Rebilder Agent Readability Spec (ARS) 0.3.0, published at rebilder.com/spec/ars with a conformance corpus. The two specs share a name and are separate documents. |
| What the audit looks at | The whole site: 25 weighted checks in four groups, covering robots.txt, sitemaps, llms.txt, markdown responses and HTML quality. It reports a result from 0 to 100, a rating and a fix hint for each failed check. | One page at a time, across seven dimensions worth 100 points. ARS also checks whether the facts that kind of page needs, such as a price or opening hours, appear early and cost little to read. Each check shows its evidence and whether it is measured or heuristic. |
| Serving agents | Middleware for Next.js, SvelteKit, Nuxt and Vercel detects agents and rewrites the request to a markdown route you provide. | The gateway renders markdown from the product, service and page data you connect. It mounts in Next.js, Node, Cloudflare Workers, Shopify and other fetch-based frameworks. |
| Blocking training crawlers | The audit’s robots.txt check fails when robots.txt blocks GPTBot, ClaudeBot, CCBot or Google-Extended from the whole site. | ARS gives a site that blocks training crawlers but allows AI assistants the same result as a fully open site. |
| Tracking | An onDetection callback runs for each detected agent request, so you can send it to your own analytics. | The Console reports AI visits, the pages agents request and the searches that found no answer. |
| After the audit | Each failed check comes with a suggested fix you can paste into a coding agent. | Customer checks and Opportunities show which answers are missing. You publish sourced answers through Answers Hub, and Autopilot keeps supported answers current on Pro. |
| Price | Free and open source. | The gateway and scanner are free. Solo is $49/mo and Pro is $99/mo. |
Where they overlap
Both help an AI agent get a clean version of your page.
Markdown at the same URL
Both answer a recognized AI agent, or a request for markdown, with markdown at the page’s own address. Both add Vary: Accept so caches keep the HTML and markdown versions apart.
Readability checks
Both check what an agent receives. Vercel’s audit and Rebilder’s scanner each look at robots.txt, llms.txt, markdown responses, canonical URLs and structured data.
What Rebilder adds
An audit tells you what to change. Rebilder helps you make the change and see what happens next.
A clean version built from your data
The gateway renders each page from the product, service and policy records you connect, with the key facts first. You connect data you already have instead of writing a markdown copy of each page.
AI visit tracking
See which AI agents reach your site, which pages they request and which visitors AI conversations send you.
Customer checks and Opportunities
Test a customer’s buying question against your pages and see which details are missing, even before AI traffic arrives.
Sourced answers in Answers Hub
Publish explainers, buying guides and comparisons on your own site, backed by your own sources.
Autopilot
On Pro, Autopilot runs scheduled checks on selected questions and updates supported answers from your sources. Each run comes with a report.
Use them together
Example: a site that deploys through CI.
1. Audit on every deploy
Run npx @vercel/agent-readability audit with your site URL, --json and a --min-score threshold. The job fails when the result drops below it.
2. Serve the clean version
Install the Rebilder gateway and connect your sources. It answers the audit’s markdown checks for AI agents and for Accept: text/markdown, and it can generate your llms.txt. Turn on its .md URLs, frontmatter, markdown sitemap and markdown 404 for the audit’s discovery checks.
3. Check key pages
Run npx rebilder check on the pages customers rely on, with --fail-on to hold a minimum ARS grade. It names the facts each page is missing.
4. Measure and improve
Watch AI visits and unanswered requests in the Console. Then publish answers where customers need more detail.
What they do better
- It is free under the MIT license, and the audit runs with one npx command in any CI system.
- Its audit covers site-wide items that ARS does not score, such as sitemap.md, working llms.txt links, .md URLs and missing-page responses.
- If your pages already exist as markdown, for example in an MDX docs site, its middleware routes agents to them with a few lines of configuration.
When to choose them instead
Choose it on its own when your pages already exist as markdown and you want a free audit and routing layer. Add Rebilder when you want the clean version built from your business data, with tracking and answers on top.
Use the evidence to decide whether the next step is a better explanation on your site or stronger sources elsewhere.
Where these figures come from
The table and notes above describe what @vercel/agent-readability publishes about itself. Products and prices change, so check each source before you rely on a detail.
Vercel publishes @vercel/agent-readability on npm under the MIT license. The latest version was 0.7.0, released September 4, 2026.
https://www.npmjs.com/package/@vercel/agent-readability · checked 2026-09-28
The README describes agent detection by user agent, the Signature-Agent header and a request heuristic, with middleware for Next.js, SvelteKit, Nuxt and Vercel that serves markdown from a route you provide.
https://www.npmjs.com/package/@vercel/agent-readability · checked 2026-09-28
The audit CLI runs 25 weighted checks in four categories, reports a result from 0 to 100 and supports
--jsonand--min-scorefor CI.https://www.npmjs.com/package/@vercel/agent-readability · checked 2026-09-28
In the package code, the robots.txt check fails when robots.txt blocks GPTBot, ClaudeBot, CCBot or Google-Extended from the whole site.
https://www.npmjs.com/package/@vercel/agent-readability?activeTab=code · checked 2026-09-28
The README links the Agent Readability Spec to a Vercel guide. That link now redirects to “Make your site readable by AI agents”, version 1.0 from August 2026.
https://vercel.com/kb/guide/make-your-site-readable-by-ai-agents · checked 2026-09-28
Want the numbers? Run the standard on any site.
Run the free scanner to inspect a public page and its response evidence. Published comparisons include dated measurements and a rerun link; verified index listings follow their own opt-in policy.
Company names and marks identify the products we compare against. All trademarks belong to their owners, no affiliation or endorsement is implied, and we remove a mark on its owner’s request.