The verified action layer for AI.

Agents should not guess what a business supports. Every domain on wh.ax has a machine-readable set of official actions (call, book, order, return, track), published and verified by the site owner. Read them, trust the provenance, act with confidence.

One address per action

Every action has a stable address. Humans open it; agents read it.

wh.ax/@honeywillow.com/book

The @domain is the verified site; the verb is the one thing the action does. See the address format.

Two feeds, one resolver

Fetch a domain's published official actions, cacheable and read-only:

Structured JSON: https://wh.ax/@<domain>/actions.json. Each action carries its address, verb, purpose, method, risk, data_shared and destination.
llms.txt: https://wh.ax/@<domain>/llms.txt. The llms.txt convention, same actions in Markdown.

Provenance you can filter on

Every feed carries a verified flag. true means the actions were verified by the site owner (DNS-verified); false means they were auto-discovered and not yet claimed. Provenance is authorship, not a safety endorsement, but it lets an agent prefer, or require, verified-only actions.

An action contract agents can reason about

method is how it acts: link, tel, mailto, form, search, info.
risk is low, medium or high, so an agent knows when to confirm before acting.
data_shared is what an action discloses, declared up front.
destination is the canonical place the action resolves to, on the owner's own site.

Building on wh.ax?

A usage-based API and MCP server for platforms, assistants, directories and agents (verified actions at scale, with metering and signed provenance) is coming. Talk to us if you want to build on it.

claim-your-site talk-to-us →