Say something.
Start with the exact sentence or phrase you actually wrote.
How it
One sentence in.
Way too much explanation out.
whatimeant.lol takes the sentence you wrote, defines every part in context, and gives the finished explanation its own read-only link.
Over-explain it.
Each word gets defined in context, then the sentence gets summarized as a whole.
Make it a link.
The finished explanation gets its own public, read-only URL.
Send that instead.
Send the URL. Anyone with it can read the explanation. You are now excused from further clarification.
Generated links are public to anyone who has the URL. Ordinary prepared results are stored for about 30 days.
privacy detailsworks
functions or operates as intended
usage note: apparently that wasn't enough information.
fine. the unnecessarily specific version.
The website does less than it used to.
Ordinary links now start in a private GPT. The public site only reads prepared explanations, so a visitor cannot trigger the paid generation step.
-
private GPT
Generates the lexical payload: token text, headwords, parts of speech, definitions, usage notes, and the overall summary.
-
authenticated publisher
Posts the original sentence and completed payload to
/api/make-linkwith a bearer token. -
validation + KV
The Worker authenticates the request, checks the payload against the lexical schema, and stores the accepted result in Cloudflare KV.
-
?text=...
Returns a
whatimeant.lol/?text=...URL. Opening it reads the stored result./api/make-linkdoes not call OpenAI.
the part where the first version got defensive
Creation is private.
/api/make-link requires a bearer token and rejects any request with a browser Origin header.
Payloads are checked.
The Worker validates the lexical schema before writing to KV. If the same normalized sentence already has a different payload, it returns 409 cache_conflict.
Old misses were signed.
The legacy /api/explode miss path requires an HMAC signature bound to the normalized sentence and a timestamp within five minutes.
Abuse was bounded.
That path also rate-limits by IP, caches results by SHA-256 sentence hash, and uses a short KV lock to reduce duplicate generation.
what happened to the AI bill?
Publishing an ordinary link no longer makes a model API call.
The first version could call OpenAI when a sentence was missing from the cache. Ordinary creation now takes a different route: the private GPT creates the payload, and /api/make-link only authenticates, validates, stores, and returns the URL.
The legacy signed /api/explode miss path still exists in the Worker and can still call OpenAI when a valid signature is supplied. It is not used by the normal private-GPT publishing flow.
The cheapest API request was the one the public stopped being allowed to make.