# superprompt.io
A prompt in, a superprompt out. Three tiers of lift. No account.

POST https://superprompt.io/v1/superprompt
  {"prompt": "...", "tier": "polish" | "elevate" | "scaffold",
   "context": "optional: code, notes, prior thread", "target": "optional: who runs it, e.g. claude code / chatgpt / a junior engineer",
   "stream": false}
  -> {"tier","model","superprompt","usage":{"cost_usd","tokens"}}
  Plain-text body in, plain text out: -H 'content-type: text/plain' ... ?format=text
GET  https://superprompt.io/v1/superprompt?q=<urlencoded>&tier=polish&format=text
GET  https://superprompt.io/v1/tiers     the tiers and what each one does
GET  https://superprompt.io/v1/stats     shared budget level, lifetime totals

Tiers
  polish    the same ask, made excellent: one kernel, real constraints, an output shape, a definition of done. Paste and go.
  elevate   the ask re-posed at the register where the best prior work lives: named framings, decisions bound, a verification plan.
  scaffold  a whole project brief with a strong baseline: intent, non-goals, architecture, style regimes, milestones, AGENTS.md, definition of done.

Limits
  prompt <= 24000 chars, context <= 60000 chars.
  Free use draws on a shared budget that drains continuously (429 names retry_after_s).
  Bring your own OpenRouter key to skip the budget and pick any model:
    -H 'X-OpenRouter-Key: sk-or-...'   with optional "model": "anthropic/claude-opus-4.8"
  Streaming: "stream": true or Accept: text/event-stream -> SSE of {"delta"} then {"done","usage"}.

Nothing is stored. Prompts pass through to the model and are gone.