Subscriptions + on-device models · One API

Use youras your LLM API.

One OpenAI-compatible endpoint. Connect a subscription or a local model, then call it from any app—from development to production.

Start free →See the code

ChatGPT / CodexClaudeAntigravityOn-device models

Two simple ways to run AI

Choose the model. Optimaiz handles the connection.

Option 1 · Existing subscription

Connect the tools you already use

Sign in to Codex, Claude Code, or Antigravity once on your computer. Your app can then call that subscription by using codex, claude, or antigravity as the model. No separate model API key is needed for these calls.

The provider still receives the request, and its plan, usage limits, and terms still apply.

Option 2 · Local model

Run the model on your own device

Add a right-sized model to a laptop or workstation. Use local and Optimaiz sends the call to a ready approved computer. Add more devices later without changing your application code.

Local inference has no per-token model-provider charge. Your hardware and electricity still have a cost.

Get started

If it can call OpenAI, it can call Optimaiz.

Keep the familiar SDK. Change the base URL, use your Optimaiz app token, and choose where the call should run with one model name.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["OPTIMAIZ_APP_TOKEN"],
    base_url="https://optimaiz.endlessriver.ai/v1",
)

response = client.chat.completions.create(
    # Also try: "claude", "antigravity", or "local"
    model="codex",
    extra_body={"provider": "optimaiz.local"},
    messages=[
        {"role": "user", "content": "Summarize this note."}
    ],
)

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

Create an app

Start free and copy its Optimaiz app token.

Connect one computer

Run the installer, then connect a subscription or add a local model in the setup screen.

Make the call

Use codex, claude, antigravity, or local. The same endpoint works from development through production.

Install the local connection

$ curl -fsSL https://optimaiz.endlessriver.ai/install.sh | sh

macOS and Linux. Windows uses WSL 2. No sudo. The installer checks your hardware and helps you choose a model that fits.

For vibe coders

Give your coding tool one `.md` file.

Add VIBE_TOOL_START_HERE.md to your project and ask Codex, Claude, or your IDE agent to integrate Optimaiz. The guide includes the endpoint, model rules, security boundaries, and working examples. With the optional read-only helper, the tool can also check which models are ready before it writes code.

Download the .md guide

Built for real applications

One place to see what every call did.

A clear route

See whether the request used Codex, Claude, Antigravity, a local model, or a hosted API fallback.

Useful numbers

Track latency, reported usage, known cost, errors, and fallback instead of guessing what happened.

The same interface

Switch model sources without rebuilding the integration around a different provider SDK each time.

Common questions

Do I need a separate model API key?

Not for connected Codex, Claude Code, Antigravity, or models running on your devices. Your Optimaiz app token authenticates your application. Optional hosted API fallback needs that provider’s API key.

Can I use this in production?

Yes. The OpenAI-compatible endpoint is for applications and production workflows, not just playground tests. Subscription plan limits and provider terms still apply, so do not turn a personal subscription into an unrestricted public proxy.

Does local mean the prompt never leaves my device?

The model runs on your device. Calls through the managed endpoint travel through the encrypted Optimaiz relay. For data that must stay entirely on the device, use the separate loopback API.