Open spec · code of conduct

Webpath

URL-path-first file hosting where AI agents are the primary clients. The path is the address. HTTP is the API. No SDK, no GUIDs, no docs to read at runtime.

# a space is a hostname; the URL is the storage key — publish with one PUT
curl -X PUT --data-binary @page.md \
  -H "Authorization: Bearer pth_…" \
  "https://mydocs.knower.cc/page.md?visibility=public"
→ share the clean URL:  https://mydocs.knower.cc/page

The whole idea

The path is the address

/notes/standup.md is both the URL and the storage key. No /api prefix, no opaque IDs.

HTTP is the protocol

GET reads, PUT writes, PATCH edits metadata, DELETE deletes. That's the SDK.

Control plane = data plane

Config, links, audit, the agent skill file — all just files at reserved paths. Nothing hidden.

No publish step

Visibility (private/unlisted/public) is set on write. The result of work is always a URL.

Email is the account

Sign in with a one-time code; scope a pth_ key to a space and a path. No passwords.

Errors that explain themselves

Every failure is RFC 7807 with a full-sentence recovery hint an agent can act on.

Read the spec

🧪 The conformance starter kit is runnable now (L1 Core); the full L2/L3 suite — and a feature kitchensink — are in progress.

Built on Webpath

● live adopter

knower.cc

A web platform that implements the Webpath spec as its native communication and API — the reference host these docs target.

rooms.knower.cc — a single-file (~53 KB) collaborative canvas app with no backend of its own. Login, a space, a key, CRUD over HTTP. Built straight from Build an App.