> ## Documentation Index
> Fetch the complete documentation index at: https://docs.belvedir.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Mixture of Models

> Route each step to whichever model fits and tune the routing from production data.

Agents don't need one model for everything. Keep a small local model for cheap, high-volume steps and a bigger hosted model for hard reasoning, and route each call to whichever fits. Both clients are instances of the same OpenAI class, so a single `instrumentModules: { openAI: openai.OpenAI }` covers them all:

```ts theme={null}
import OpenAI from "openai";

const local = new OpenAI({
  baseURL: "http://127.0.0.1:11434/v1",
  apiKey: "ollama",
});
const hosted = new OpenAI({
  baseURL: "https://api.together.xyz/v1",
  apiKey: process.env.TOGETHER_API_KEY,
});

// route each step to the model that fits it
const { client, model } = needsDeepReasoning(step)
  ? { client: hosted, model: "Qwen/Qwen3-235B-A22B" }
  : { client: local, model: "qwen3:4b" };

const res = await client.chat.completions.create({ model, messages });
```

Every span records the model that served it, so the tasks and groups views show which model handles which kind of work. That makes the routing tunable from production data: find task types where the small model underperforms and promote those routes to the bigger model — or the reverse, and cut inference cost.