The open protocol that makes websites readable by AI — 10x faster, at 1/100th the cost.
The web was designed for humans in 1993. AI agents today face six fundamental frictions that CBOR-Web eliminates:
| # | Problem | The cost | CBOR-Web fix |
|---|---|---|---|
| 💸 | Token waste | Agents pay for 93% noise (CSS, ads, trackers). ~$36/day/agent | 95% signal ratio. 100x cheaper. |
| 🧩 | Chaotic parsing | 1,400 DOM elements per page. Agents guess what is content | Typed blocks (h, p, table). Zero ambiguity |
| 🔍 | Blind discovery | No standard to tell AI "here's my full site" | /.well-known/cbor-web. Entire site, one request |
| 🪪 | No identity | Publishers don't know which agent reads their site | Cryptographic wallet per agent. Access control |
| 🤑 | Economic plunder | Content scraped for free. Zero revenue for publishers | Token-gated premium. Agent pays to read |
| 🏛️ | Legal blind spot | Digital identity law incoming. Nobody is compliant | Identity built in. Day-1 compliance |
Every day, millions of AI agents — ChatGPT, Claude, Gemini, Perplexity, enterprise RAG pipelines — crawl the web to extract information. They all face the same problem: HTML was designed for humans in 1993.
| HTML today | CBOR-Web |
|---|---|
| Downloads 1.6 MB of menus, scripts, ads, trackers | Reads 50 KB of pure structured content |
| 93% of what an agent downloads is noise | 95% signal ratio |
| Costs $3 to crawl a 100-page site | Costs $0.01 |
| No standard way for sites to talk to machines | Open protocol, IETF standard (RFC 8949) |
The web has two clients now. Only one of them is served.
CBOR-Web lets any website publish a machine-readable copy of its content — in parallel with its existing HTML. Humans see the regular site. AI agents get a binary, structured, 10x smaller version.
It's like having a REST API for your content, automatically generated from your HTML. No CMS change. No redesign. One command.
text2cbor --input ./my-site --output ./cbor --domain mysite.comCBOR-Web is built on three strategic pillars aligned with the biggest shifts in AI and digital regulation.
No other protocol lets a webmaster know which AI agent is accessing their site. With CBOR-Web, every agent carries a cryptographic wallet identity. The publisher can:
- Allow or block specific agents
- Set access tiers (public L0 vs token-gated L1)
- Audit exactly which agent read which page, when
This is critical as the French digital identity regulation extends to AI agents. The upcoming law will require all automated agents accessing web content to be identifiable. CBOR-Web is already compliant — your site will be, too, without any additional work.
Every AI agent crawling the web wastes 93% of the tokens it pays for just stripping HTML noise. For inference alone, a single agent reading 1000 pages/day wastes ~$36/day. At scale — millions of agents — this represents billions of tokens wasted daily, including during fine-tuning where clean training data is critical.
CBOR-Web reduces token consumption by 100x. An agent reading a CBOR-Web site pays only for the content — not the markup. For AI companies training foundation models, clean structured data means higher-quality fine-tuning at a fraction of the cost.
France is advancing legislation on digital identity for automated systems. Every AI agent accessing public web services will need to prove its identity. CBOR-Web is the only protocol with native cryptographic identity built in — from day one.
Your site adopting CBOR-Web today means you are pre-compliant with the regulation before it even passes.
| Metric | Value |
|---|---|
| Sites live | 38 domains, 59 pages across 3 languages |
| Edge CDN | Cloudflare global network (300+ cities) |
| SDK languages | 8 (TypeScript, React, Python, Go, PHP, Ruby, Java, C++) |
| IETF alignment | Built on RFC 8949 (CBOR), RFC 8610 (CDDL), RFC 8615 (Well-Known URIs) |
| npm package | @deltopide_edy/cborweb — zero dependencies |
Your site already has an HTML version. Adding CBOR-Web takes 30 minutes:
- Install the converter:
cargo install text2cbor - Point it at your HTML:
text2cbor generate -i ./html -o ./cbor -d mysite.com - Serve the CBOR directory alongside your HTML
Benefits:
- AI agents prefer your site over competitors (lower cost = more crawls)
- You get analytics on which agents visit, what they read
- You control access levels — public content vs premium content
- Zero impact on your existing site. Humans see zero difference.
npm install @deltopide_edy/cborwebimport { CBORWebClient } from "@deltopide_edy/cborweb";
const site = new CBORWebClient("https://cbor.deltopide.com");
// Read the entire site in one request
const manifest = await site.manifest();
// → { name: "Deltopide", pages: 49, languages: ["fr", "en", "es"] }
// Or fetch a single page — already structured
const about = await site.page("/a-propos");
// → { title: "À propos", blocks: [{type: "h", text: "..."}, {type: "p", text: "..."}] }Same API available in Python, Go, PHP, Ruby, Java, C++ — zero dependencies.
| robots.txt | sitemap.xml | llms.txt | Schema.org | CBOR-Web | |
|---|---|---|---|---|---|
| Content delivered | No | No | Summary | Metadata | Full structured content |
| Binary efficiency | — | — | — | — | Zero tokenization cost |
| Access control | No | No | No | No | Token-gated premium |
| Agent analytics | No | No | No | No | Doléance feedback protocol |
| Edge delivery | — | — | — | — | Cloudflare CDN |
| Multi-language | — | — | — | — | Per-page, native |
CBOR-Web doesn't replace these standards — it's the final layer that gives AI agents what they actually need: the content.
Your HTML site ──→ text2cbor ──→ .cbor files
│
▼
AI Agent ──→ Cloudflare Edge ──→ cbor-server ──→ Structured binary content
(300+ cities) (Rust/axum) (CBOR, RFC 8949)
The full specification is documented in CBOR-WEB-CORE.md and companion documents.
git clone https://github.com/ploteddie-bit/cbor-web.git
cd cbor-web/tools/text2cbor
cargo build --release
./target/release/text2cbor generate -i ./my-html-site -o ./cbor-output -d mysite.comnpm install @deltopide_edy/cborwebOr pick your language: Python · Go · PHP · Ruby · Java · C++ · React
- Dashboard: cbor-web.explodev.workers.dev — 38 sites, live manifest
- Spec: draft-plot-cbor-web-00.md — IETF Internet-Draft
- Core protocol & reference implementation (Rust)
- 8-language SDK ecosystem
- Cloudflare global edge deployment
- 38-site production proving ground
- IETF Internet-Draft submission
- npm package publication
- WordPress plugin (43% of the web)
- Shopify app (e-commerce integration)
- IETF Working Group adoption
- SECURITY.md — vulnerability reporting and production checklist
- DoS protection: max 50MB input, depth 50, 100K items per structure
- CORS: read-only (GET/HEAD) on edge worker, safe header forwarding only
- CI:
cargo auditblocks builds on critical vulnerabilities - Secret scanning: GitHub push protection blocks accidental credential leaks
- Depfu: automatic security updates via Dependabot
Specification: CC BY 4.0 — open standard, attribution required.
Reference implementation: MIT — use freely, contribute back.
"The web has two clients. It's time to serve both."
CBOR-Web — ExploDev / Deltopide SL — 2026