Backend AI

JavaScript/TypeScript native W7S backends can use W7S AI without adding provider credentials to the repo. W7S injects these bindings during deploy:

• W7S_AI: internal service binding to the W7S AI runner.
• W7S_AI_TOKEN: W7S-generated secret token for this deployment.

No w7s.json entry, W7S account, provider account, provider API token, card, or GitHub secret is required for hosted deploys that use this binding.

Run a Model

backend/index.ts

type Env = {
  W7S_AI: Fetcher;
  W7S_AI_TOKEN: string;
};

export default {
  async fetch(_request: Request, env: Env) {
    const response = await env.W7S_AI.fetch("https://w7s.internal/api/v1/ai/run", {
      method: "POST",
      headers: {
        authorization: `Bearer ${env.W7S_AI_TOKEN}`,
        "content-type": "application/json"
      },
      body: JSON.stringify({
        model: "@w7s/meta/llama-3.1-8b-instruct-fp8",
        input: {
          prompt: "Write one short, original deployment joke."
        }
      })
    });

    return response;
  }
};

The response is JSON:

{
  "status": "success",
  "data": {
    "model": "@w7s/meta/llama-3.1-8b-instruct-fp8",
    "result": {
      "response": "..."
    }
  }
}

Limits

W7S resolves the caller from the deployment token, checks app suspension, applies ai.run usage limits, and then calls the platform-owned AI binding. The first supported model is @w7s/meta/llama-3.1-8b-instruct-fp8; more models can be enabled by the W7S platform operator.

See w7s-io/example-ai-joke for a full frontend plus backend example.