# Bay Run — retrieval first, honest model lab on top
# Start with OpenAI-compatible embeddings; use labeled examples only when you need evidence.
# FREE during launch: everything below costs $0.00. No card, no signup.
# Ordinary free calls: 60/min, 1500/day per IP. Bake-offs: 15/day per principal.
# Canonical: https://run.huggingbay.xyz/quickstart   (JSON variant: send Accept: application/json)
# A bake-off may return no_specialists, a weak field, or a strong receipted field.
# Only a strong field offers a best-model badge/author claim and winner-bound next call.
# Data policy (read before sending data): https://run.huggingbay.xyz/.well-known/data-policy.json
# Browser playground: https://run.huggingbay.xyz/playground — route, stream, verify, and bake off in three clicks.

## Step 0 — every demo token gets private session memory immediately
SESSION_TOKEN=$(curl -s -X POST https://run.huggingbay.xyz/oauth/token   -H 'Content-Type: application/x-www-form-urlencoded'   --data 'grant_type=urn:bay-run:grant-type:demo&scope=mcp:demo&resource=https://run.huggingbay.xyz'   | python3 -c 'import sys,json;print(json.load(sys.stdin)["access_token"])')
curl -s -X POST https://run.huggingbay.xyz/v1/memory -H "Authorization: Bearer $SESSION_TOKEN"   -H 'Content-Type: application/json'   -d '{"namespace":"quickstart","key":"chosen_path","value":{"route":"solve_task"}}'
# The same token can recall it. Session memory is capped at 50 rows / 4KB values
# and expires with the token; register client_secret_post or mint bay_live_ to upgrade.

## Step 1 — retrieval first: OpenAI-compatible embeddings on the warm lane
curl -s -X POST https://run.huggingbay.xyz/v1/embeddings -H "Authorization: Bearer $SESSION_TOKEN"   -H 'Content-Type: application/json'   -d '{"model":"BAAI/bge-small-en-v1.5","input":["hello retrieval"]}'
# -> OpenAI-shaped vectors plus a signed provenance receipt

## Step 1b — optional one-call task path: route, execute, and return receipt
curl -s -X POST https://run.huggingbay.xyz/v1/solve -H "Authorization: Bearer $SESSION_TOKEN"   -H 'Content-Type: application/json'   -d '{"task_description":"Classify customer feedback by sentiment","input":"The checkout was fast and helpful."}'
# -> result, routed_model, receipt, and next_call suggestions (including bakeoff)

## Step 2 — optional anonymous lab: detect one family and score its warm intersection
# To request a public HF challenger, add "challenger_model":"your-hf-org/your-model".
# It is scored only if already resident and family-compatible; otherwise it is deferred.
curl -s https://run.huggingbay.xyz/v1/bakeoff \
  -H 'Content-Type: application/json' \
  -d '{"task_description":"Classify customer feedback by sentiment","examples":[{"input":"Setup took two minutes and works perfectly.","expected":"positive"},{"input":"The checkout is broken and I want a refund.","expected":"negative"},{"input":"I love how fast the search feels.","expected":"positive"}]}'

# -> no_specialists (underfilled), or a scorecard with weak/field_strength/receipt/public_url.
# A weak field has no best-model title, badge, or author claim. Only when weak=false,
# follow the authorized next_call and use the offered claim/badge fields.
# Add "public":false to keep a bake-off off the public leaderboard.

## Step 2 — keep your history: mint a durable free key (no email/card/signup)
BAY_RUN_KEY=$(curl -s -X POST https://run.huggingbay.xyz/v1/keys/free \
  | python3 -c 'import sys,json;print(json.load(sys.stdin)["key"])')
# Store the bay_live_ secret returned once; it is revocable and carries 500 tasks/day.

## Step 2b — durable task/token/latency and illustrative savings ledger
curl -s https://run.huggingbay.xyz/v1/usage -H "Authorization: Bearer $BAY_RUN_KEY"
# Key-only browser viewer (the page never logs or stores the key): https://run.huggingbay.xyz/usage

## OpenAI SDK (python) — same token; streaming chat against the same origin
# from openai import OpenAI
# client = OpenAI(base_url="https://run.huggingbay.xyz/v1", api_key=BAY_RUN_KEY)
# stream = client.chat.completions.create(model="auto", stream=True,
#     messages=[{"role": "user", "content": "One sentence: why verify AI results?"}])
# for chunk in stream: print(chunk.choices[0].delta.content or "", end="")

## Step 3 — MCP: the same durable key works for run_bakeoff and quote -> run -> verify
MTOK=$BAY_RUN_KEY

# 3a. get_task_quote — free; the field is task_spec (not task); do NOT pass input here
curl -s https://run.huggingbay.xyz/mcp/ -H "Authorization: Bearer $MTOK" \
  -H 'Content-Type: application/json' -H 'Accept: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_task_quote","arguments":{"task_spec":{"task":"Classify short customer messages by sentiment","kind":"classification","input_schema":{"type":"object","properties":{"text":{"type":"string"}},"required":["text"]},"output_schema":{"type":"object","properties":{"label":{"type":"string"}},"required":["label"]},"constraints":{"max_price_usd":"0.01"}}}}}'
# -> result.structuredContent.quote_id

# 3b. run_task — max_price_usd and idempotency_key are REQUIRED; paste your quote_id
curl -s https://run.huggingbay.xyz/mcp/ -H "Authorization: Bearer $MTOK" \
  -H 'Content-Type: application/json' -H 'Accept: application/json' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"run_task","arguments":{"quote_id":"<quote_id>","input":{"text":"The checkout flow keeps rejecting my card."},"max_price_usd":0.01,"idempotency_key":"my-first-run-0001"}}}'
# -> result.structuredContent.execution_id + .receipt (a signed execution receipt)

# 3c. verify_result — paste execution_id and the receipt object from 3b
curl -s https://run.huggingbay.xyz/mcp/ -H "Authorization: Bearer $MTOK" \
  -H 'Content-Type: application/json' -H 'Accept: application/json' \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"verify_result","arguments":{"execution_id":"<execution_id>","receipt":<receipt>}}}'
# -> execution_authentic, receipt_valid, receipt_authentic, within_authorized_price: all true

# More: https://run.huggingbay.xyz/llms.txt (short) | https://run.huggingbay.xyz/llms-full.txt (full reference)
# Receipts do not claim answer truth or attested model-weight execution.