AIQR 爱码
Agent Ready QR — 让二维码不仅能被手机识别,更能被 AI 理解。
A QR code today carries a URL. A phone knows what to do with it. An AI agent does not — and, more to the point, an agent usually cannot decode the QR at all. Vision models read pictures; they do not run a QR decoder. A QR code is an opaque black square to them.
AIQR does two things, and deliberately nothing else:
1 · A caption an agent can read without decoding anything
┌─────────────────────┐
│ ███████ │ ← ordinary QR, untouched
│ ██ QR ██ │
│ ███████ │
└─────────────────────┘
─────────────────────────
Midea air conditioner KFR-35GW ← what it is. Works with no network at all.
https://id.gs1.org/01/06901010101010 ← where the contract lives
AIQR:0ATA-XJH3-YXYC-PCC5 ← sentinel + self-repairing identifier
Printing human-readable characters under a code is fifty years old — that is what the digits under a UPC barcode are. What is new is that the first line carries meaning, the third line carries a machine sentinel, and all three are aimed at a vision model rather than a laser scanner.
2 · A vocabulary for what an agent may do
Four link relation types, carried inside an ordinary RFC 9264 linkset:
| Term | Meaning |
|---|---|
aiqr:mcp | A Model Context Protocol server through which an agent can operate this entity. |
aiqr:action | One action with business meaning that an agent may perform on this entity. |
aiqr:safety | Constraints an agent must respect when operating this entity. |
aiqr:knowledge | A document an agent can read to answer questions about this entity. |
aiqr:actionId | A stable, unique identifier for an action. Used as an RFC 9264 extension attribute on an aiqr:action link. |
What AIQR does not do
It does not define a document format, a discovery mechanism, a registry, or a new barcode. Those exist and are deployed:
| Layer | Standard AIQR uses |
|---|---|
| Barcode | ISO/IEC 18004 (QR) |
| Identifier in the code | GS1 Digital Link |
| Discovery | GS1-Conformant Resolver |
| Contract format | RFC 9264 linkset |
| Document link types | GS1 Web Vocabulary |
| Caption typeface | ISO 1073-2 (OCR-B) |
| Tool protocol | Model Context Protocol |
A vendor already running a GS1 resolver adopts AIQR by adding link entries. Not by hosting anything new, and not by registering with anyone — including this domain.
Status
Draft. The identifier string is stable at v1.1; the caption block and vocabulary are v0.2. The core assumption — that a vision model reads the caption but cannot decode the QR — has been measured, not assumed. So has the identifier's resistance to OCR error. Both measurements, including what they fail to prove, are in the repository.
Documents
| Document | |
|---|---|
| AIQR 规范 v0.2 | 字幕块与 Agent 词汇表。§0 记录了 v0.2 砍掉了什么、为什么。 |
| AIQR 识别字符串规范 v1.1 | 字符集、格式、GF(17) 校验、解析要求。已稳定。 |
| OCR 实测 | 2,900+ 次识别。找到一个真实缺陷,推翻两条推理。 |
| 决定性测试 | 核心假设:视觉模型解不出二维码,但读得懂字幕。 |
All CC0. Every quantitative claim in the specifications is produced by a script in the reference implementation, including the claims about what the measurements fail to prove.