7-Day Money-Back Guarantee
Try ProxyLLM risk-free. See Terms
Codex Hosted for AI agencies
Scale your AI agency without worrying about AI costs.
ProxyLLM runs OpenAI's Codex for you, signed in with your ChatGPT account. Your agents and automations bill to the flat plan you already pay for instead of per-token API pricing.
terminal
# point your existing OpenAI SDK at ProxyLLM
export OPENAI_BASE_URL="https://api.proxyllm.ai/v1"
# requests now run through Codex on your ChatGPT subscription.
# your prompts and code stay exactly the same. Works with
What Codex Hosted does.
Connect an account, we run Codex for you, your whole stack calls one endpoint.
01 · Connect
Sign in with your ChatGPT account.
OpenAI's official device-code login, the same flow Codex uses on your laptop.
Click connect, approve the code at chatgpt.com, done. We never see your password. The session lives in a container only your account uses, and you can disconnect it from the dashboard at any time.
device login openai.com
$ codex login --device-auth
visit chatgpt.com/codex/device
enter code XQZ4-JKPT
Codex connected · container started
02 · Run
We run Codex for you, around the clock.
Your container runs codex exec, the CLI's documented non-interactive mode.
Every request your apps send becomes a Codex run inside your private container. The work bills to your ChatGPT plan, usage resets the way your plan resets, and the dashboard shows what each request would have cost at API rates.
codex exec · isolated container live
req_4281 done billed to subscription
req_4282 done billed to subscription
req_4283 done billed to subscription
plan usage 38% resets in 3h 12m
03 · Call
Your whole stack calls one endpoint.
OpenAI-compatible. Set a base URL, keep your code.
Node, Python, curl, n8n, LangChain: anything that accepts an OpenAI base URL works. When you hit your plan's limit, requests fall back to a second connected account or your own API key until it resets.
one endpoint fallback order
OPENAI_BASE_URL="https://api.proxyllm.ai/v1"
1 · Codex · account A active
2 · Codex · account B fallback
3 · OpenAI API key fallback
Codex Hosted
One flat plan instead of a metered bill.
Connect your ChatGPT subscription once. ProxyLLM runs Codex in a container only you use, and your workloads stop billing per token. The dashboard shows exactly how much work the plan absorbed.
- Connect with OpenAI's official device-code login
- Isolated container per account, never shared
- Falls back to a second account or your API key at the limit
June statement
Codex connected
ChatGPT Pro 5x
Covers up to $3,500 of API-equivalent usage
$100.00
ProxyLLM SaaS
Hosted Codex, keys, logs, dashboard
$129.00
Total for $3,500 of work
Instead of paying direct API rates
$229.00
Savings calculator
Put in your current OpenAI bill.
This uses your bill amount only. No token estimates, no guesswork. ProxyLLM maps your existing API spend to the Codex Hosted subscription tier that can absorb it.
ChatGPT Plus capacity $700 API-equivalent / mo
ChatGPT Pro 5x capacity $3,500 API-equivalent / mo
ChatGPT Pro 20x capacity $14,000 API-equivalent / mo
Recommended setup
ChatGPT Pro 5x capacity
Covers up to $3,500 of API-equivalent work each month.
Codex Hosted
Current OpenAI bill
$3,500
ProxyLLM + ChatGPT
$229
Monthly savings
$3,271
ProxyLLM SaaS fee $129/mo
ChatGPT Pro 5x subscription $100/mo
Example: a $3,500 OpenAI bill becomes $229/mo, so you save $3,271.
Compare ProxyLLM to the API and to DIY.
Same Codex, same models. The difference is what a unit of work costs you and what you have to maintain yourself.
Codex Hosted
ProxyLLM
$129/mo
OpenAI API direct
Per-token pricing
Codex CLI on your laptop
Free, manual, one machine
What usage costs you
Your flat ChatGPT plan
Per token, every call
Your flat ChatGPT plan
Works as an API endpoint
Yes, OpenAI-compatible
Yes
No
Serves apps and automations
Yes
Yes
Scripts on one machine
Runs with your laptop closed
Yes
Yes
No
At the plan limit
Second account or API-key fallback
Bill keeps growing
Wait for the reset
Multiple ChatGPT accounts
Pooled with fallback
Not applicable
One login at a time
Request logs and cost dashboard
Built in
Usage page only
None
Setup and upkeep
Sign in once
None
You maintain it
Best for
Production workloads on subscription capacity
Spiky, low-volume use
Personal terminal use
Built for teams whose API bill is the problem.
If you ship AI features for clients and watch the margin disappear into per-token billing, Codex Hosted is the fix.
ProxyLLM is for you if
- Your OpenAI bill runs hundreds to thousands a month
- Your workloads look like agency work: content, agents, automations, code
- You already pay for ChatGPT and want it doing production work
- You want one endpoint your whole stack can point at
- You want a fallback at the plan limit, not an outage
Probably not for you if
- You make a handful of AI calls a day, total
- Every call needs sub-second latency
- Compliance requires direct provider contracts
- You are pre-launch with no live AI traffic yet
Connect once. Keep your SDK.
ProxyLLM is OpenAI-compatible. Connect your ChatGPT account, point your existing code at our base URL, and your requests run through your hosted Codex.
01
Connect your ChatGPT account
Sign in with OpenAI's device-code flow. Your private Codex container starts in about a minute. Add a second account or an API key as fallback.
02
Point your SDK
Set OPENAI_BASE_URL to https://api.proxyllm.ai/v1 and use your ProxyLLM key. Your existing code keeps working.
03
Watch the API bill drop
Work that used to bill per token now runs on your subscription. The dashboard shows what each request would have cost at API rates.
One price, the whole suite. No inference markup.
Try ProxyLLM risk-free
ProxyLLM is a suite of developer tools for working with AI providers. One membership covers gateway, routing, sub-keys, Blitz, schema outputs, analytics, and Codex Hosted.
Membership
$129 per month
- Codex Hosted: run production AI traffic on your ChatGPT subscription instead of paying for tokens. Far cheaper than API costs.
- One endpoint for OpenAI, Anthropic, and OpenRouter with your own keys
- Auto-routing picks the cheapest model that can still handle each prompt
- Send batches of prompts in one call, with spending limits per project
- Dollar-cost dashboard for every request, day, and model
- Telegram alerts when usage spikes or a fallback fires
7-Day Money-Back Guarantee
Try ProxyLLM risk-free. See Terms
Common questions
What is Codex Hosted?
ProxyLLM runs OpenAI's Codex on our servers, signed in to your ChatGPT account. Your apps call one OpenAI-compatible endpoint, each request runs as a Codex job inside your private container, and the work bills to your flat ChatGPT subscription instead of per-token API pricing.
Is this safe to use?
Yes. The Codex terms allow programmatic usage: codex exec is the CLI's documented non-interactive mode, built for scripts and pipelines, and Codex is included in ChatGPT plans. We run the official, unmodified CLI in a container that belongs to you alone. You sign in directly with OpenAI through the CLI's built-in device-code flow, the same sign-in OpenAI documents for headless machines, and we never see your password.
Is this against OpenAI's terms?
Programmatic use is intended functionality: OpenAI documents codex exec for automation and recommends signing in with your ChatGPT account. We built around the account rules: one user, one container, one account, your own workloads, never shared or pooled. OpenAI still has the final call. Its terms let it restrict accounts at its discretion, and we are not affiliated with or endorsed by OpenAI. If OpenAI's posture changes, we tell you immediately and comply.
What happens when I hit my limit?
Requests keep flowing. Connect more than one Codex account and the next one takes over as fallback, or set your existing OpenAI API key as the fallback until your plan's limit resets. The dashboard shows which lane served each request.
Can I use Claude Code instead of Codex?
No. Claude Code is not supported for this kind of programmatic usage, so we don't support it. Anthropic's docs state it does not permit third-party services to route requests through Claude subscription credentials, and it cut off tools that tried in January 2026. We won't put your Claude account at risk. Codex is the lane whose vendor documents the use.
How much does it save?
Depends on your bill. A $3,500/month OpenAI API bill fits in ChatGPT Pro 5x at $100 plus ProxyLLM at $129, about $229 total. Put your own number in the calculator and see which plan tier absorbs it.
How does connecting my account work?
You click connect, OpenAI shows you a code, and you approve it at chatgpt.com. That is OpenAI's official device-code login, the same flow the Codex CLI uses on your laptop. We never see your password, and the session lives only inside your isolated container.
What happens to my API keys?
Encrypted with AES-256-GCM in our database. Decrypted only inside the serverless function that calls the provider on your behalf. We never log them and never send them to anyone other than the provider you pointed us at.
Will my stack work with it?
If it accepts an OpenAI base URL, yes. Official OpenAI SDKs, n8n, LangChain, Cursor, plain curl: set OPENAI_BASE_URL to our endpoint, use your ProxyLLM key, and keep your code.
Why $129/month?
Because hosting your Codex container, keys, logs, and dashboard is software, and we do not mark up inference. If the subscription lane saves you more than $129 a month, it pays for itself. A $3,500 bill saves over $3,200.
What if OpenAI changes its policy?
Then we change with it. We built the feature so it can be adjusted or removed at OpenAI's direction without touching your data or the rest of the product. Your keys, logs, and dashboard stay yours either way.