Skip to main content

Getting Started — Yabanin Operator (Prepaid Credits)

Yabanin is self-hosted: you operate the gateway, the admin panel and the billing relationship with your users. There is no hosted sign-up and no price list in this repository — provider prices and any resale markup are configured in your own database. This guide walks an operator through a working prepaid setup.

1. Deploy the stack​

make docker-up        # postgres + redis + migrations + gateway + admin
# or a single process:
yabanin dev --migrate

The admin panel listens on :8086 (compose host :9091). Create your administrator account with yabanin admin create-user (see --help; the CLI refuses inline passwords — use --password-stdin or YABANIN_ADMIN_PASSWORD_FILE).

2. Configure prices​

Model prices live in the models_pricing table and are what the gateway bills against. Load them from your providers' price sheets:

yb pricing sync --apply     # sync provider prices into models_pricing

After a sync the gateway picks up database prices within 5 minutes (DBPricer cache). Whether you resell at a markup is your commercial decision; Yabanin can compute resale amounts from a configured markup, but the rate is set in your deployment, not published here.

3. Create an org and an API key​

In the admin panel: create the customer organization, then mint an API key for it (api_keys). Keys are org-scoped; spend is attributed through labels and session keys.

4. Top up prepaid credits​

Yabanin bills prepaid: requests are only served while the org holds credits.

# Admin API (admin session auth; cross-org views are root-only):
POST /api/orgs/{org_id}/credits/topup          # add credits
GET  /api/orgs/{org_id}/credits/balance        # current balance
GET  /api/orgs/{org_id}/credits/transactions   # ledger

When the balance is exhausted the gateway rejects inference requests with 402 Payment Required (insufficient_credits) until the next top-up. GET /api/credits/low-balance lists orgs approaching zero. How you collect payment (invoice, Stripe, bank transfer) is outside Yabanin's scope.

5. First request​

export YABANIN_API_KEY="<key you minted>"
export BASE_URL="http://localhost:8082/v1"

curl "$BASE_URL/chat/completions" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $YABANIN_API_KEY" \
  -d '{
    "model": "gpt-4o",
    "messages": [{"role": "user", "content": "Hello, World!"}],
    "max_tokens": 50
  }'

The response is standard OpenAI chat format (plus SSE with "stream": true). Spend is recorded server-side — the gateway does not add cost fields to the chat response. To read spend:

  • per session: GET /v1/yabanin/session-keys/{id}/usage (returns cost_usd);
  • per org / per label: the admin analytics and the request log (request_logs, see analytics.md).

Where to look next​