@aiqlick/inference — TypeScript client

A thin wrapper over the official openai package for the AIQLick inference API.

Deliberately thin. The whole product claim is that the API is OpenAI-compatible, so a client that reimplemented the protocol could drift away from that claim without anyone noticing. AIQLick extends OpenAI, so anything the official client does — streaming, tool calling, retries, whatever it adds next — works here unchanged.

Install

npm install @aiqlick/inference openai

openai is a peer dependency, so you control its version and there is never a second copy of the SDK in your tree.

Use

import { AIQLick, Models } from "@aiqlick/inference";

const client = new AIQLick({ apiKey: process.env.AIQLICK_API_KEY });

const reply = await client.chat.completions.create({
  model: Models.CHAT_DEFAULT,
  messages: [{ role: "user", content: "hello" }],
});
console.log(reply.choices[0]?.message?.content);

Streaming works exactly as it does with the official client:

const stream = await client.chat.completions.create({
  model: Models.CHAT_DEFAULT,
  messages: [{ role: "user", content: "hello" }],
  stream: true,
});
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}

Models

Use the aliases, not provider model ids:

| Constant | Alias | |---|---| | Models.CHAT_PREMIUM | aiqlick/chat-premium | | Models.CHAT_DEFAULT | aiqlick/chat-default | | Models.CHAT_CHEAP | aiqlick/chat-cheap | | Models.EXTRACT_DEFAULT | aiqlick/extract-default | | Models.EXTRACT_FAST | aiqlick/extract-fast | | Models.EMBED_DEFAULT | aiqlick/embed-default |

The mapping from an alias to a specific provider model is a platform decision that changes without callers changing — that is the entire point of the gateway. Pinning a provider id in your own code gives that up.

Your key may be scoped to a subset. client.models.list() returns only what it can actually call.

Errors

The OpenAI SDK gives you a status and a message. asAIQLickError() promotes our code onto a typed error, which is what lets you tell "out of credits" from "subscription lapsed" — the same 402 with entirely different fixes.

import { asAIQLickError, ErrorCodes } from "@aiqlick/inference";

try {
  await client.chat.completions.create({ /* … */ });
} catch (err) {
  const specific = asAIQLickError(err);
  if (specific?.code === ErrorCodes.INSUFFICIENT_CREDITS) {
    // top up or upgrade the plan
  }
  throw specific ?? err;
}

It returns undefined for anything it does not recognise rather than guessing, so a provider error never gets mislabelled as a platform error.

| code | Meaning | |---|---| | INSUFFICIENT_CREDITS | Credit balance exhausted | | SUBSCRIPTION_INACTIVE | No active subscription for the company | | LLM_MODEL_NOT_ALLOWED | Key not permitted to use that alias | | LLM_GATEWAY_UNAVAILABLE | Gateway unreachable — retry with backoff |

Configuration

| Variable | Default | |---|---| | AIQLICK_API_KEY | — | | AIQLICK_BASE_URL | https://api.aiqlick.com/llm/v1 |

Reading the base URL from the environment means one build can be pointed at dev or a self-hosted deployment without a code change.

Not using this client

You do not have to. The API is OpenAI-compatible, so the official client works directly — this package only saves you the base URL and the constants:

import OpenAI from "openai";
const client = new OpenAI({
  apiKey: "sk-…",
  baseURL: "https://api.aiqlick.com/llm/v1",
});

Full documentation: https://docs.aiqlick.com