TanStack Start

@mynthio/sdk/tanstack-start exports mynthWebhookHandler for a Start server route. Generation from a server function uses the normal SDK. If the app already uses TanStack AI's generateImage(), the separate @mynthio/tanstack-ai-adapter package plugs Mynth into it.

Receive the webhook

npm install @mynthio/sdk

Put MYNTH_WEBHOOK_SECRET, the whsec_ secret from registering the endpoint, in the server environment.

// src/routes/api/mynth-webhook.ts
import { createFileRoute } from "@tanstack/react-router";
import { mynthWebhookHandler } from "@mynthio/sdk/tanstack-start";

export const Route = createFileRoute("/api/mynth-webhook")({
  server: {
    handlers: {
      POST: mynthWebhookHandler({
        imageTaskCompleted: async (result, { request }) => {
          console.log(result.taskId, result.images, request.url);
        },
        imageTaskFailed: async (result) => {
          console.error(result.taskId, result.errors);
        },
      }),
    },
  },
});

The handler verifies the Standard Webhooks webhook-signature against webhook-id, webhook-timestamp and the raw body, and requires the body's id to be the webhook-id. A missing or wrong signature, or a timestamp more than five minutes off, answers 400. An event with no handler, or one it does not know, answers 200. It reads the secret on each request, from MYNTH_WEBHOOK_SECRET or from { webhookSecret } as the second argument, and throws if neither is set.

It reads a clone of the request, so the route's request stays unread. Each callback receives a mapped result or task failure and the route context: request, params, and middleware context, plus the verified event. event.id is the webhook-id, the same on every retry, so store it to skip events you have already handled. The handler names are the same as in every SDK helper. See the table on webhooks. Unsigned per-request webhook.custom deliveries are rejected unless you pass { unsigned: { verify } }. See per-request webhooks.

Generate on the server

Keep MYNTH_API_KEY on the server and call the SDK from a server function:

import Mynth from "@mynthio/sdk";

const mynth = new Mynth();

const pending = await mynth.image.generateAsync({
  model: "black-forest-labs/flux.2-pro",
  prompt: "A lighthouse at dusk, film grain",
  generate_public_access_token: true,
});

return { taskId: pending.id, token: pending.access.publicAccessToken };

generate() waits for the task. generateAsync() returns the id right away, plus a browser token when the request sets generate_public_access_token, for when the caller cannot wait. Tasks and polling covers both.

TanStack AI

@mynthio/tanstack-ai-adapter is for apps that already call generateImage() from @tanstack/ai. It calls the Mynth API itself, so it needs no other package: only @tanstack/ai 0.61.0 or newer as a peer. It does not verify webhooks. The route above still does that.

npm install @mynthio/tanstack-ai-adapter @tanstack/ai

import { generateImage } from "@tanstack/ai";
import { mynthImage } from "@mynthio/tanstack-ai-adapter";

const result = await generateImage({
  adapter: mynthImage("black-forest-labs/flux.2-pro"),
  prompt: "A lighthouse at dusk, film grain",
});

console.log(result.images[0]?.url);

• mynthImage(model, config?) binds one model. The key comes from MYNTH_API_KEY or config.apiKey.
• createMynthImage({ apiKey, baseURL, defaultHeaders, destination }) returns a function that takes a model, for several adapters sharing one config.
• Image content parts become Mynth inputs on models listed in MYNTH_IMAGE_INPUT_MODELS. Other Mynth fields go in modelOptions.
• MYNTH_IMAGE_MODELS is the id list built into the package version, and it includes auto. Pass an explicit id. The live list is GET https://api.mynth.io/models.
• Failures are MynthError subclasses with a code that TanStack AI reports on RUN_ERROR: the API's own code (such as insufficient_balance) for MynthAPIError, or MYNTH_TASK_FAILED, MYNTH_NO_IMAGES and MYNTH_TASK_TIMEOUT.

The adapter creates the task, polls it until it settles, and returns the finished images. tanstack-start-ai-mynth-adapter is a complete Start app that streams the TanStack call.

Next steps

• Webhooks: registering the endpoint, events, sources, and retries.
• Generate images: the request fields.