API

Vyre speaks the OpenAI Chat Completions format, so the official OpenAI SDKs work with it for text chat. It’s meant to be called from servers.

• Base URL: https://usevyre.xyz/v1
• API key: your sk_lite_… key. In the app, open Backup & recovery to copy it. It holds your credits, so treat it like a password.

Quick start

import OpenAI from 'openai';

const vyre = new OpenAI({ baseURL: 'https://usevyre.xyz/v1', apiKey: process.env.VYRE_KEY });
const r = await vyre.chat.completions.create({
  model: 'auto',
  messages: [{ role: 'user', content: 'Summarize this in two sentences: …' }],
});
console.log(r.choices[0].message.content);

import os
from openai import OpenAI

vyre = OpenAI(base_url="https://usevyre.xyz/v1", api_key=os.environ["VYRE_KEY"])
r = vyre.chat.completions.create(model="fast", messages=[{"role": "user", "content": "Hello"}])
print(r.choices[0].message.content)

Endpoints

• POST /v1/chat/completions: streaming with stream: true; function tools, response_format, temperature and other common parameters are supported. max_tokens defaults to 4096, up to 8192.
• GET /v1/models and GET /v1/models/{id}

Models

auto, fast, kimi, glm, coder, best, deliberate and image. See pricing for what each is good at and what it costs.

Vyre extras

• web: "auto" lets the model search the web when needed; "always" searches before every reply. Sources arrive as url_citation annotations.
• thinking: "quick" | "balanced" | "deep" sets how much the model reasons first.
• redact: true hides names, emails, phone, card and wallet numbers from the model and restores them in the reply. With it, our server sees that request’s text; the model sees only placeholders.

Billing and errors

Each request reserves its worst-case cost and settles to the real cost. Errors use OpenAI’s shape: 401 for an unknown key, 402 for not enough credits, 404 for an unknown model, 429 for rate or daily limits (30 requests a minute per key), and 502 when the model service fails, which isn’t charged.