Agent-facing export
GET /api/notes/exportreturns the whole table. Ask for it withAccept: application/tronand the payload becomes self-describing TRON — declarations first, then positional rows — which an LLM tool or any third party can read cold at a fraction of the tokens. The route also caches the encoded body per Accept variant for 60 s.
From the terminal
curl -s https://norns-demo.humansynthesis.ai/api/notes/export | head -c 300 curl -s -H 'Accept: application/tron' https://norns-demo.humansynthesis.ai/api/notes/export | head -20 curl -sI -H 'Accept: application/tron' https://norns-demo.humansynthesis.ai/api/notes/export | grep -iE 'etag|cache'
- The TRON body starts with
class C0: id,title,body,created_at,updated_atandtable C0: $— enough for a reader that has never seen the contract. - JSON and TRON are separate cache entries (
Vary: accept); a second fetch within a minute answersx-norns-cache: hiton a deployed Worker (locally the proxy's Cache API is a no-op), andIf-None-Matchgets a 304 anywhere. - Token counts here are a chars ÷ 4 estimate; the measured range on tabular data is 26–61% fewer.