@runbit/model

The model interface an agent talks to. Providers implement it; agents never import a provider.

npm install @runbit/model

import { defineModel } from "@runbit/model";
import type { ChatModel } from "@runbit/model";

// Wrap any chat function into a ChatModel that emits llm:request / llm:response.
const echo = defineModel("echo", async ({ messages }) => ({
  text: messages.at(-1)?.content ?? "",
  usage: { inputTokens: 0, outputTokens: 0 },
}));

• ChatModel — { name, chat(input, ctx) }.
• ChatInput / ChatOutput — messages + tool specs in; text, toolCalls, usage out.
• defineModel(name, fn) — adds event emission (including token usage) for free.

Real providers: @runbit/openai.

Part of Runbit — the AI Runtime. Everything executes through one small, event-driven runtime; agents, workflows, RAG, and observability are packages on top.

• Umbrella: @runbit/runtime
• License: MIT