Scalix AI

Scalix AI models behind one OpenAI-compatible API

Overview

Scalix AI provides LLM inference through an OpenAI-compatible API. The Scalix Lumio model family — from the low-latency Lumio Lite up to the Candela flagship — sits behind one endpoint, with unified authentication, smart routing with automatic fallback, cost controls, and usage tracking.

Quick Start

bash

scalix-cloud ai infer "Explain quantum computing in one sentence" --model scalix-lumio-lite

The prompt is a positional argument; pass the model with --model.

REST API

The API is OpenAI-compatible — existing code works with a URL change:

bash

curl -X POST https://api.scalix.world/v1/ai/chat/completions \
  -H "Authorization: Bearer $SCALIX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "scalix-lumio-lite",
    "messages": [
      {"role": "user", "content": "What is Scalix Cloud?"}
    ]
  }'

Streaming

bash

curl -X POST https://api.scalix.world/v1/ai/chat/completions \
  -H "Authorization: Bearer $SCALIX_API_KEY" \
  -d '{
    "model": "scalix-lumio-lite",
    "messages": [{"role": "user", "content": "Write a haiku"}],
    "stream": true
  }'

SDK

The request/response envelope is OpenAI-compatible; the SDK exposes it as the chatCompletion function:

typescript

import { chatCompletion } from "@scalix-world/sdk";
 
const { data, error } = await chatCompletion({
  body: {
    model: "scalix-lumio-lite",
    messages: [{ role: "user", content: "Hello!" }],
  },
});
 
if (data) console.log(data.choices[0].message.content);

Streaming

streamChatCompletion posts to /v1/ai/chat/completions/stream and the response is server-sent events — pass parseAs: "stream" to consume the raw stream (see the TypeScript SDK guide for details):

typescript

import { streamChatCompletion } from "@scalix-world/sdk";
 
const { response } = await streamChatCompletion({
  body: {
    model: "scalix-lumio-lite",
    messages: [{ role: "user", content: "Write a story" }],
  },
  parseAs: "stream",
});

Available Models

List the live catalog at any time:

bash

scalix-cloud ai models

Model	Best for	Availability
`scalix-lumio-lite`	Fast, lightweight responses for chat and high-volume tasks — the default model	Available
`scalix-lumio-pro`	Balanced quality for production apps	Coming soon
`scalix-candela`	Most capable flagship model for the hardest problems	Coming soon

scalix-lumio-lite is generally available and is the default — requests that omit model use it. scalix-lumio-pro and scalix-candela are announced and coming soon; they are not callable yet. Embeddings run as a built-in platform service (powering ScalixNova vector search, Enclave, and semantic features) — every plan includes them; there is no separate embedding model to select.

All AI usage is billed per-token from your credit pool. See Billing for per-model rates.

Features

Tool/Function calling: Full support for OpenAI-compatible tool use
Vision: Multi-modal inputs (images + text)
Streaming: Token-by-token streaming with metering
Cost controls: Per-key spending limits and model allowlists
Usage tracking: Token counts and cost attribution per request