Resources
One SDK: server billing, client UI, model tracking.
@crediball/sdk wraps your actions and bills credits. @crediball/react and @crediball/elements render the UI. Nothing gets priced in code — the dashboard owns that.
Install
@crediball/sdk — server-side billing
Initialize once with your secret key, then wrap every action worth billing. Crediball auto-discovers new actions; you price and activate them from the dashboard, not in code.
import crediball from "@crediball/sdk";
crediball.init(process.env.CREDIBALL_API_KEY!);
const poem = await crediball.action("generate_poem", userId, () =>
generatePoem(topic)
);No result to wrap? Use crediball.track(action, userId)instead — same billing, no return value. When you’re done wrapping, run npx @crediball/sdk sync once so every action shows up on the dashboard immediately, ready to price, instead of only after a user triggers it.
@crediball/sdk/ai — model usage tracking
Wrap your model client — wrapOpenAI, wrapAnthropic, wrapGemini, or crediballMiddleware for the Vercel AI SDK — and call it inside a crediball.action()callback. Provider, model, and token usage show up automatically on your dashboard’s Models page, without any pricing table or model manifest in your code.
@crediball/react — drop-in credit UI
One provider, three tiers of components, all reading the same live data: fully-styled drop-ins (<CreditsBadge/>, <PaywallModal/>, <ActionCostList/>), small themeable variables you place inline (<CrediballBalance/>, <CrediballTopUpButton/>), and a headless useCrediball() hook for full control. Not using React? @crediball/elements ships the same three tiers as framework-agnostic web components.
MCP server
Crediball also runs an MCP server at /api/mcp, so an AI coding agent can finish setup in the same conversation as the integration: list discovered actions, price and activate them, create top-up packages, set a signup bonus, and flip the app to live — all without a human clicking through the dashboard.
Full reference
The complete, always-current integration guide — including every method, the sandbox vs. live model, and the model-attribution rules — lives at crediball.ai/llms.txt, written for both humans and coding agents to read directly.
Get started