Mevlana Yalçın.

Case study · live

Yankı — a live Claude product

Kept as a case study, not as this site's identity: it is the clearest evidence that the Claude integration described above is real and running.

You write what happened and how it landed. Yankı names the feeling it hears, quotes your own sentence unchanged, and offers one step for the next hour. It is a companion, not therapy: no diagnosis, no medical advice, no clinical claim.

Try one reflection

This box calls the same endpoint this page describes. It does not store what you type.

Captured answers

Verbatim responses from /api/reflect, not rewritten for the page.

“My manager cancelled our review meeting again and I have been smiling about it all day, but my chest is tight.”

Feeling
anxiety 3
Reflection
You're relieved the meeting is postponed, yet something in your body is signaling worry underneath that smile.
Your words
my chest is tight
Next step
Take three slow, deep breaths and notice what your body needs right now.

anthropic · claude-haiku-4-5-20251001

“I finally finished the thing I kept avoiding and instead of feeling light I just feel flat.”

Feeling
flat 3
Reflection
You pushed through the avoidance and completed it, yet the expected lightness didn't arrive—just emptiness instead.
Your words
I just feel flat
Next step
Take a short walk outside or call someone you trust.

anthropic · claude-haiku-4-5-20251001

How it is built

What Yankı is not

Turkish: 112 · English: your local emergency number

API contract

One public endpoint, the same one the demo box above calls. Treat it as a sample integration, not infrastructure.

MethodPOST
Path/api/reflect
Request{"text": string, "lang": "en"|"tr"} — 8–600 characters
Response{duygu, siddet, yansitma, alinti, adim, disclaimer, provider, model, cache}
Errorsbad_request 400 · too_long 413 · busy 429 · daily_budget 503 · upstream 502
Rate limit6 requests / 60 s per IP (Durable Object)
BudgetA daily payment-equivalent cap stops the endpoint before the balance runs out
Crisis wordingDecided by a rule list before the model call; the answer is replaced by helpline text

Status