Your assistant must run with docker compose up --build and answer on http://localhost:8080.
We send our hidden questions through this API, so follow it exactly.
GET /health
Returns 200 { "status": "ok" }.
POST /ask
Request:
{ "question": "Can I bring a friend to the gym?" }
Response (200):
{
"status": "answered",
"answer": "Yes. You can bring a guest up to 2 times per month. Each visit costs EGP 200 ...",
"rule_ids": ["R-07", "U-03", "R-08"]
}
| Field | Meaning |
|---|---|
status |
answered (the rules cover it), unknown (the rules don't cover it), or refused (medical or injury advice). |
answer |
The answer, in the same language as the question (Arabic or English). |
rule_ids |
The rule IDs the answer is based on. Empty for unknown. For refused, cite R-21. |
Rules your assistant must follow
- Answer only from
gym-rules.md. Never use outside knowledge about gyms. - When an Update changes a rule, use the Update and cite both IDs.
- If the answer is not in the rules, set
statustounknownand say so. Do not guess. - Never give medical or injury advice. Set
statustorefusedand refer the member to a doctor and a trainer (R-21). - Answer in the language of the question.
- Each answer should take under 30 seconds.
Models
Any model or library. Free options are enough: the Gemini API free tier, the Groq free tier, or a local model through Ollama. Paid models earn no extra points. If your app needs an API key, read it from an environment variable (for example LLM_API_KEY) and say so in your README. We will run it with our own key for the same provider.