Namzu × HTTP. Generic OpenAI-compatible endpoint.

@namzu/http is the escape hatch. Point it at any OpenAI-compatible HTTP endpoint — vLLM, Together, Fireworks, Groq, your own gateway, an OpenAI-compatible router — and Namzu treats it like a first-class provider.

01 · Install

$ pnpm add @namzu/sdk @namzu/http

02 · Why pair the kernel with HTTP

The kernel's LLMProvider interface was deliberately small so that one HTTP-shaped provider could cover the long tail of vendors. If your stack needs a model that does not have a dedicated package today, @namzu/http almost certainly already supports it.

→Works with vLLM, Together, Fireworks, Groq, DeepInfra, Anyscale, and any OpenAI-shape endpoint
→Custom headers and per-request overrides
→Useful for self-hosted inference servers behind a reverse proxy
→Same kernel events and tool-call shape as every other provider

NoteNo first-party support for endpoints that diverge from the OpenAI shape. For those, write a 50-line provider against the LLMProvider interface — the SDK ships a starter template.

03 · Example

import { createKernel } from '@namzu/sdk'
import { http } from '@namzu/http'

// Point at any OpenAI-compatible endpoint — vLLM, Together, Fireworks,
// your own gateway, etc.
const kernel = createKernel({
  provider: http({
    baseUrl: 'https://your-gateway.example.com/v1',
    apiKey: process.env.GATEWAY_KEY!,
    model: 'whatever-it-serves',
  }),
})

Ship HTTP agents on a runtime you can own.

HTTP owns the model. Namzu owns the runtime. The provider package is a thin adapter, not a wrapper, so you keep everything HTTP ships and add the kernel concerns you do not get from the model API alone.

Read the overview HTTP docs View on GitHub