Framework Quickstart

Router Core is not yet published to a registry. Run from dari-router-core/ in the Dari monorepo.

Try The Example

Install and run

bun install --frozen-lockfile
bun run example

No API key needed. It routes by prompt length using an in-memory executor.

Make a real provider call

OPENAI_API_KEY=... bun run example:pi

This runs through the built-in Pi runtime with a real OpenAI model.

Create A Server

import { createPiRuntime, createRouter } from "@dari/router-core";

const apiKey = process.env.OPENAI_API_KEY;
if (!apiKey) throw new Error("OPENAI_API_KEY is required");

const pi = await createPiRuntime({ apiKey });
const router = createRouter({
  runtime: pi,
  models: [
    pi.model("openai/gpt-5.4-mini"),
    pi.model("openai/gpt-5.4"),
  ],
  policy: ({ request, candidates }) => {
    const promptLength = JSON.stringify(request.items).length;
    const preferred = promptLength > 500
      ? "openai/gpt-5.4"
      : "openai/gpt-5.4-mini";
    return {
      model: candidates.find((c) => c.id === preferred)?.id
        ?? candidates[0]!.id,
      reason: "Route longer requests to the larger model.",
    };
  },
});

Bun.serve({ port: 3000, fetch: router.fetch });

Send a request:

OpenAI
Anthropic

curl http://localhost:3000/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "my-router",
    "messages": [{"role": "user", "content": "Explain cache locality."}]
  }'

curl http://localhost:3000/v1/messages \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "my-router",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Explain cache locality."}]
  }'

The response includes the chosen model in dari_routing and the X-Router-Selected-Model header. The incoming model is a caller-facing router name — it does not force the policy to pick that model.

Use From A Sibling Project

Build the package

(cd ../dari-router-core && bun install --frozen-lockfile && bun run build)

Add the dependency

{ "dependencies": { "@dari/router-core": "file:../dari-router-core" } }

bun install

The package emits ESM in dist/. It requires Bun or Node.js 22.19+ and uses web-standard Request, Response, and AbortSignal. Bun can serve router.fetch directly; Node needs a Request/Response adapter.

Next Steps

Models

Declare candidates and eligibility.

Policies

Write routing policies.

Pi Runtime

Credentials for multiple providers.

Custom Executors

Integrate another SDK.

Protocols And Streaming

Request and streaming details.

Get Started

Build A Router

Advanced

Framework Quickstart

Try The Example

Create A Server

Use From A Sibling Project

Next Steps

Models

Policies

Pi Runtime

Custom Executors

Protocols And Streaming

​Try The Example

​Create A Server

​Use From A Sibling Project

​Next Steps

Models

Policies

Pi Runtime

Custom Executors

Protocols And Streaming

Try The Example

Create A Server

Use From A Sibling Project

Next Steps