Skip to content
Neutrome docs

Your first model

Create a namespace before creating a model. In Compute → Models, create a model with code support, select its namespace, then replace the starter source with this executor.

first-model.ts
import type { Executor } from "@neutrome/lilsdk";


const upstream = "default/gpt-oss-20b";


const executor: Executor = {
  async execute(request, ctx) {
    return ctx.invoke(upstream, request);
  },


  async *stream(request, ctx) {
    yield* ctx.invokeStream(upstream, request);
  },
};


export default executor;

ctx.invoke() serves non-streaming calls. ctx.invokeStream() serves streaming calls. Both receive the normalized request program and invoke the target model or provider through the managed runtime.

default/gpt-oss-20b is a built-in managed model, so this first model works without provider setup. Save, deploy, then test the model from the editor.

Later, replace it with an ID from the built-in provider catalog or <provider-code>/<provider-model> after you connect an upstream provider.

Model editor showing TypeScript executor source

For executor types and runtime context details, use the SDK reference.