Codex Hosted: 25 Questions, Answered Directly

These are the 25 questions we actually get about Codex Hosted, each answered in one to three sentences. The shortest version of all of them: we run OpenAI’s official Codex CLI on our servers under your own ChatGPT account and expose it as an OpenAI-compatible endpoint, so programmatic work bills to your flat plan instead of the token meter.

Where an answer deserves a full page, the link follows it.

The basics

What is Codex Hosted?

We run OpenAI’s official, unmodified Codex CLI in a container that belongs to your account alone, signed in with your ChatGPT credentials, and expose it as an OpenAI-compatible endpoint. Requests run as Codex usage and bill to your flat plan. The full definition lives in what is Codex Hosted?

Is ProxyLLM affiliated with OpenAI?

No. ProxyLLM is independent software, not affiliated with, endorsed by, or partnered with OpenAI. We run the CLI exactly as OpenAI ships it.

Do I need an OpenAI API key?

No. The Codex lane runs entirely on your ChatGPT account. An API key is optional and acts as a metered fallback lane for overflow.

Which ChatGPT plans include Codex?

All of them: Free, Go, Plus, Pro, Business, Edu, and Enterprise. Paid plans carry the windows that matter for real workloads, and OpenAI publishes current limits on its pricing page.

Setup and sign-in

How long does setup take?

About five minutes, with nothing to install: device-code sign-in, a container start, and one base URL change. Commands for every step are in the setup guide.

How does the ChatGPT sign-in work?

Through OpenAI’s documented device-code flow. You approve a code at chatgpt.com while signed in as yourself, and OpenAI issues the session directly into your container.

Does ProxyLLM see my password?

Never. The sign-in happens between you and OpenAI, and the device-code exchange never shows us a password to see or store.

What do I change in my app?

Two environment variables, nothing else:

export OPENAI_BASE_URL="https://api.proxyllm.ai/v1"
export OPENAI_API_KEY="pllm_your_key_here"

Can I disconnect my account?

Yes, anytime, from the dashboard. Disconnecting ends your container’s session; your ChatGPT subscription with OpenAI continues untouched.

Compatibility

Will my existing OpenAI SDK code work?

Yes. Anything that accepts an OpenAI base URL works unchanged: official SDKs, n8n, LangChain, plain curl. Request and response payloads are standard OpenAI shape.

Does the Codex lane stream responses?

No. The Codex lane returns complete responses; API-key fallback lanes stream normally. Backend jobs rarely notice, and chat UIs should plan around it.

Which models can I call?

The models Codex currently serves. Ask the endpoint itself with GET /models rather than trusting a blog post, because the surface follows what OpenAI ships to Codex.

Can I run embeddings or fine-tunes?

Not on the Codex lane; those are API-platform features. Keep them on your OpenAI key, direct or as a passthrough lane with no markup.

Is it a fit for chat interfaces?

Only if the interface tolerates complete responses. Token-by-token rendering belongs on an API-key lane, which streams.

Limits and fallback

What happens when I hit my plan’s limit?

The gateway fails over: a second connected ChatGPT account, then your OpenAI API key, then back to the subscription lane when the window resets. Full behavior in what happens when you hit your Codex limit.

How many requests does a plan include?

OpenAI varies limits by plan and model and adjusts them over time, so we never hardcode counts. Current figures live on OpenAI’s Codex pricing page, and the dashboard shows your observed usage against the window.

Can I connect two ChatGPT accounts?

Yes, if you own both. Each gets its own isolated container and its own window, and you set the fallback order.

How do I know which lane served a request?

The request log names the lane on every entry, so subscription work and metered overflow are visible per call rather than on the invoice.

Billing

What does ProxyLLM charge?

$129 per month for Codex Hosted, flat. No inference markup, no per-request fee, no percentage of your spend.

Is there a free tier?

Yes. Starter is $0: bring your own keys as a passthrough lane, with request logs and the dashboard. The Codex lane is what the $129 buys.

What does OpenAI bill me?

Your ChatGPT plan, directly, like any other subscriber, plus normal API rates on whatever your fallback key serves. We sit in the middle of neither charge.

How much API-equivalent work does a plan absorb?

Our planning estimates: Plus at $20 about $700 a month, Pro 5x at $100 about $3,500, Pro 20x at $200 about $14,000. Estimates, never guarantees; the calculator maps your bill to a tier.

Policy

Is Codex Hosted against OpenAI’s terms?

Programmatic Codex use is documented functionality, Codex is included in ChatGPT plans, and your account is never shared or pooled. OpenAI’s terms still govern your account, and OpenAI has the final call. The full reading is in is Codex Hosted against OpenAI’s terms?

Can OpenAI restrict my account anyway?

Yes. OpenAI retains discretion over its services, which is why we built inside the documented account rules; if OpenAI directs a change, we comply immediately, as our terms state formally.

Can I do this with a Claude subscription?

No. Anthropic does not permit third parties to route requests through Claude plan credentials and enforced that position in January 2026, so this approach is specific to OpenAI’s Codex.

That is the whole catalog. If your real question was “what would my bill become”, it has a precise answer: put your current API spend into the calculator and read the result.