A while back we shared how we'd started writing a separate document just to explain our products to AI agents. Then we noticed the same problem, one level up.
An agent arrives at iwasoft.com. How is it supposed to know what this place is, what to read, what it can actually do here, and what it's allowed to do? It has to guess again — crawling pages and trying to infer the answer from what it finds.
A single file
This time the fix was even simpler: one file at
/.well-known/ai-manifest.json. It points the agent to the right
places — our content (llms.txt), our machine docs, our API — and,
most importantly, it states plainly what AI is and isn't allowed to do.
It doesn't compete with the standards that already exist. It points to them:
- read — where the machine-readable content lives (llms.txt, sitemap)
- docs — the machine-first product docs (our AI Docs Standard)
- act — how an agent may take action (MCP, OpenAPI, or declared actions)
- policy — the one thing no other standard covers: what agents are permitted to do
If the first standard was “explain the product to the agent,” this one is “explain the site to the agent.” They're siblings.
It's open
Both are open (CC BY 4.0) and live on iwasoft.com as reference implementations. You can adopt the AI Manifest in three steps: publish the file, point it at what you already have, and state your AI-use policy.
- GitHub: github.com/iwasoftcom/ai-manifest-standard
- Write-up: iwasoft.com/standards/ai-manifest
- This site's live manifest: /.well-known/ai-manifest.json