# AX-SYS-01: Make your purpose machine-readable

> Context and comprehension · foundational · AX-SYS-01

An agent that cannot parse what your product does cannot use it or cite it. Expose your purpose through structured data, semantic HTML, an llms.txt index, and hosted context files.

## Why

Stripping HTML chrome to markdown cuts agent tokens 80–90% and improves retrieval (Cloudflare/Fern analysis). The canonical AX framework (agentexperience.ax — Contextual Alignment) requires that services actively provide context because agents must not be assumed to have sufficient internal context or a correctly trained model. Biilmann identifies this as the Context pillar: "does the LLM know about your product?"

## Do

Ship `/llms.txt`, JSON-LD structured data, and clear `<meta>` descriptions. Also provide hosted context files agents can use directly: AGENTS.md (coding agent rules), Agent Skills files, and Arazzo workflow specs for complex multi-step interactions. See agentexperience.ax for the full context file ecosystem.

## Don't

Rely on JS-rendered pages with no structured data, or assume the model was trained on current documentation about your product.

## Artifact

`/llms.txt`:

```
# Acme
> One-line description of what Acme does.
## Docs
- [Quickstart](https://acme.com/docs/quickstart.md): get running in 5 minutes
- [API reference](https://acme.com/docs/api.md): endpoints and schemas
```

Pair with JSON-LD on key pages:

```html
<script type="application/ld+json">
{"@context":"https://schema.org","@type":"SoftwareApplication","name":"Acme",
 "description":"Acme helps teams do X.","url":"https://acme.com",
 "offers":{"@type":"Offer","price":"29","priceCurrency":"USD"}}
</script>
```

## Citations

1. [agentexperience.ax — Principles of AX](https://agentexperience.ax/concepts/principles-of-ax/) — Contextual Alignment: digital services must not assume an agent has sufficient internal context; provide structured context via llms.txt, MCP, agents.json, Arazzo, and AGENTS.md
2. [Mathias Biilmann — AX in Practice](https://biilmann.blog/articles/ax-in-practice/) — LLMs need precise, structured documentation about what is special about your product, including unique APIs and knowledge the model may not have from training
3. [Cloudflare / Fern — Token Analysis](https://buildwithfern.com/post/optimizing-api-docs-ai-agents-llms-txt-guide) — Stripping HTML chrome to markdown cuts agent tokens 80–90% and improves retrieval accuracy
4. [llms.txt standard — Jeremy Howard / Answer.AI](https://llmstxt.org/) — Proposed September 2024; standardised way for websites to provide key information to LLMs via a root-level index file

---
Source: Agent Experience Principles (axprinciples.com)