ckl75/semantic-manifest-specification — explained in plain English
Analysis updated 2026-05-18
Publish a machine readable map of your site's content for AI crawlers to read efficiently.
Let language models discover relationships between pages without crawling each one individually.
Pair a page with its plain text markdown twin for easier text ingestion.
Adopt the spec's reference format to speed up how fast AI bots can crawl a large site.
| ckl75/semantic-manifest-specification | 0-bingwu-0/live-interpreter | 010zx00x1/faresnipe | |
|---|---|---|---|
| Stars | 2 | 2 | 2 |
| Language | — | Python | Python |
| Setup difficulty | moderate | moderate | easy |
| Complexity | 3/5 | 2/5 | 2/5 |
| Audience | developer | general | general |
Figures from each repo's GitHub metadata at analysis time.
Requires generating and hosting an NDJSON manifest file for your own site.
The Semantic Manifest Specification is an open data standard for describing a website's content and structure in a format that AI crawlers, language models, and retrieval systems can read efficiently. It defines a streamable file format called NDJSON, where each line is its own small piece of JSON, so a crawler can read a site's content types, how pages relate to each other, and which pages have plain text markdown versions, one line at a time without loading everything into memory at once. The project argues that existing web standards were not built for this job. Sitemaps only list raw page addresses without saying how those pages relate to each other. JSON-LD only describes one page at a time and cannot show the shape of an entire site. Plain text files meant to guide AI crawlers use up a lot of a language model's limited context space once a site has thousands of pages. The Semantic Manifest is meant to close that gap by giving crawlers a compact, line by line map of the whole site. The repository includes the written specification itself, version 0.1, along with a reference example file showing what a real manifest looks like. According to the project, after linking a manifest on a roughly 58,000 page site, an AI crawler called ClaudeBot was able to read the entire site at about seven pages per second, and a demo video of that crawl is linked in the project. The specification is already used as a built in part of a larger content publishing system called the High-Velocity Content Engine, and two live example manifests from real sites are linked for reference. The specification and its reference file are released under the CC0 public domain license, meaning anyone can use them freely without restriction.
An open NDJSON based data standard that lets AI crawlers and language models read a whole site's content map efficiently.
Setup difficulty is rated moderate, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.