Skip to content
DocsOverviewMenu

Documentation

Build on a private, OpenAI-compatible API

Lirux serves open-weight models for chat, embeddings, speech and reranking behind OpenAI-compatible endpoints for supported API patterns. If your code already uses an OpenAI SDK, integration is a base URL, an API key and a model name.

About API access

API access is issued during onboarding. Endpoints and parameters on these pages describe the production API; contact us if you need something not listed.

At a glance

Base URL

https://api.lirux.ai/v1

Authentication

Authorization: Bearer sk-…

Example model

qwen3-30b-a3b
from openai import OpenAI
import os

client = OpenAI(
    base_url="https://api.lirux.ai/v1",
    api_key=os.environ["LIRUX_API_KEY"],
)

response = client.chat.completions.create(
    model="qwen3-30b-a3b",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Summarise our Q3 support tickets."},
    ],
)

print(response.choices[0].message.content)

Explore the docs

QuickstartInstall the OpenAI SDK, set two environment variables and send your first request.AuthenticationBearer tokens, key rotation, scoping and keeping keys out of client code.API referenceEvery endpoint with parameters, example requests and example responses.ModelsHow model IDs work, what is available on shared and dedicated capacity.ErrorsError object format, status codes and a retry strategy that respects limits.OpenAI compatibilityWhat is supported, what is partial and what is not — in one matrix.

Endpoints

Versioning and stability

The API version is part of the path: every endpoint lives under /v1. Within a version we only make additive changes — new endpoints, new optional parameters, new response fields and new models. Your client should ignore fields it does not recognise.

A change that would break existing integrations ships under a new version path, and the previous version keeps working during a migration period agreed with customers. Model IDs are versioned separately; see pinning model versions.

Conventions

  • All requests use HTTPS. Request and response bodies are JSON (application/json), except audio uploads, which use multipart/form-data.
  • Timestamps are Unix seconds. Token counts are reported by the model's own tokenizer.
  • Every response carries an x-request-id header. Include it when you contact support about a specific request.
  • Rate-limit state is returned in x-ratelimit-* headers — see rate limits.

Getting help

Questions about an integration, a parameter you need, or a model you want to run: write to [email protected] or talk to an engineer. Moving an existing OpenAI integration? Start with Migrate from OpenAI.

Already integrated? The /v1/models endpoint always shows which model IDs your key can call.