jev-seo audits any site from one homepage URL, checks 52 Google Search Central rules and asks a small model for meaning

An SEO shop or a founder about to review a client site can now run a full audit from one homepage URL and get a ranked list of fixes back for one cent in model calls, on the author's own numbers. jev-seo, released on 22 September 2026 under an MIT licence by developer Agrici Daniel, has passed 139 stars and 32 forks in four days on GitHub.

The tool crawls a site, checks it against 52 rules tied to Google Search Central, measures Core Web Vitals, and asks Jev, TypeSafe's System One decision model, typed questions about every page it finds. Version 0.1.1 runs as a Claude Code skill using the command /jev-seo https://example.com or from a plain shell. The output is one PDF for the client, one XLSX action tracker for the work, and one Markdown report, all built from a single audit.json.

What it checks

Agrici Daniel lists the 52 rules in eight groups on the repository: crawl and indexing, on-page factors, content quality, links, structured data (including Google's required rich-result properties), AI crawler access, performance and security. Every check runs against the live site.

On top of the rules, the tool sends the pages to Jev with typed questions about page type, search intent, importance, helpfulness, specificity, trust, citability, and how well the title and meta match the content. It also flags pages competing with each other for the same searches. The readme states each answer keeps its full probability distribution, so a low-confidence call carries its own low probability into the report where a reader can see it.

What it costs to run

The standard mode uses Jev for the language part and needs no third-party SEO data subscription. The readme puts the model spend at "about a cent in Jev per site". An optional --full mode adds rankings, keywords and backlinks from DataForSEO, which the readme prices at "about 0.30 USD" per site. Every figure here is the author's, on the sites they tested, and a shop running many sites will need to time and price its own runs.

What lands in the report

Each recommendation the tool writes has an ID, a priority, an impact score, an effort estimate, the evidence that triggered it, a proposed fix and a link to an official source. Heuristics are labelled as heuristics. The readme states that the written summary at the top of the PDF is checked against the underlying audit before the file renders, so the paragraph a client reads matches the table beneath it.

The traction

Agrici Daniel's account was created in July 2025, has 2,993 followers and 72 public repositories. The jev-seo repository is a week old and already sits at 139 stars with 32 forks. The code is Python and the licence is MIT; the bundled fonts carry the SIL Open Font License. A CI badge on the readme shows the checks passing on the main branch as of the last commit.

An SEO audit that runs from one URL, cites an official source on every action, and costs a cent in model calls fits a team about to write a proposal for a new client. The audit output is long, so the team still has to decide which fixes to make first. The Jev judgements come from a small classifier model, so treat those answers as a first read that a human confirms. Run the tool against a site the team already knows to calibrate the output before pointing it at a paying client.

Source

• Repository: AgriciDaniel/jev-seo on GitHub