If you already pay for a ChatGPT plan, you can plug that same Codex subscription straight into Hermes and stop bouncing between five different apps to get work done. Once it is connected, the same subscription powers regular chat, parallel sub-agents that draft multiple blogs at once, ChatGPT Images 2.0, delegated coding tasks, and image analysis, all inside one desktop workbench.
This walkthrough breaks down exactly how to log in, how the pricing and model selection works, and how to put Codex to work across writing, design, development, and vision. By the end you will see how a marketer can write three blogs, generate cover images, analyze a screenshot, and build a working app in about half an hour without ever leaving Hermes. Here is exactly how to set it up.
Core Concepts
- One subscription, every task: your existing Codex (ChatGPT) plan runs chat, sub-agents, images, coding, and vision inside Hermes.
- No API keys: connecting is a one-click browser login and authorize flow, not a copy-paste key setup.
- Session-based pricing: Codex usage runs on five-hour session limits plus a weekly cap that scales with your plan tier.
- Parallel sub-agents: launch several agents at once to draft multiple pieces of content or handle separate tasks simultaneously.
- ChatGPT Images 2.0 built in: generate and edit images at the same quality you would get inside ChatGPT, without switching apps.
- Delegated coding: hand a build off to Codex as the runtime or orchestrator and get a working app, repo, and tests back.
- Vision analysis: drop in a screenshot or photo and have Codex describe and analyze it.
Who Does This Apply To
This is for marketers, solo operators, and small agencies who already pay for ChatGPT and want more leverage from that subscription. If you are tired of jumping between ChatGPT, a separate coding tool, and your notes app, and you want a single command center where AI drafts content, builds tools, and handles visuals in the background while you do other work, this workflow is built for you.
What Running Codex Inside Hermes Actually Means
Running Codex inside Hermes means your ChatGPT subscription becomes the engine behind a single desktop app that handles almost everything. Instead of opening ChatGPT for writing, a separate environment for coding, and yet another tool for images, you select Codex as your model in Hermes and route all of it through one interface.
Hermes is an open-source AI agent that works in both a desktop app and a terminal. The desktop app, in particular, turns into a kind of workbench: you can keep multiple sessions running side by side, watch sub-agents work in real time, and switch between writing, design, and development without losing your place. If you want the broader picture on the platform itself, the rundown on the Hermes desktop app that no longer needs a terminal is a good companion read.
Expert tip: The desktop app is the easiest place to set this up. You can launch it from a terminal session by typing hermes desktop, but once it is open you rarely need to touch the command line again.
Connecting Your Codex Subscription to Hermes
Connecting takes about a minute and never asks you for an API key. Open the Hermes desktop app, go to Settings in the top right, and open the Providers section. Your available providers load there, and you simply click to connect OpenAI / Codex.
Here is the full flow:
- Open the Hermes desktop app (or launch it from a terminal with
hermes desktop). - Go to Settings in the top right corner.
- Open the Providers tab and let it load.
- Click to connect OpenAI / Codex. A verification page opens in your browser.
- Log in and authorize, just like any other login screen.
- You are connected. Hermes now uses whatever ChatGPT plan you are on (Plus, Pro, etc.) and the Codex portion of it.
That is the entire process. There is no key to paste, no config file to edit, just a browser handshake. This is the same kind of one-click connection that powers tool integrations through connecting AI with your business tools over MCP.
Codex Pricing and Model Selection, Explained
Codex pricing runs on session and weekly limits rather than per-token billing, so you are not watching a meter on every message. You get a five-hour session limit and a weekly limit, and higher-tier plans simply give you more usage to work with.
To pick Codex as your model, type codex in the model selector at the top of the chat. You will see it pull from your OpenAI Codex subscription, and you can choose a reasoning level such as minimal, fast, thinking, high, or max. Higher effort settings like Max and Fast produce stronger results but burn through your session allowance more quickly, so it pays to match the setting to the task.
From experience: You can set Codex as your default model by simply asking Hermes to do it (“set up Codex as my default model”). To check how much you have left, launch Codex and open Settings, then look at usage remaining for your five-hour session and weekly percentages. For everyday work, a balanced effort level like High is usually the sweet spot, with Max reserved for heavier jobs.
Launching Sub-Agents to Write Multiple Blogs at Once
Sub-agents let you delegate several tasks in parallel from a single prompt, and they are one of the biggest reasons to run Codex through Hermes. Instead of writing one blog, waiting, and writing the next, you brief multiple agents and they all work at the same time.
In the walkthrough, the setup was a Notion page containing three empty child pages, one for each topic: how to set up Google Ads, how to set up responsive search ads (RSAs), and how to choose locations in Google Ads. Because an AI-powered workspace like Notion was connected over MCP, Hermes could read the parent page, find the three child pages, and write directly into each one.
The flow looked like this:
- Give Hermes the link to the parent Notion page and confirm it has access.
- Prompt it to launch sub-agents to write a blog for each topic on the page.
- Hermes writes itself a script, finds the three child pages, and launches three writing sub-agents.
- Each sub-agent drafts and publishes its blog independently, and you can watch all three working in the background.
- When they finish, Hermes verifies each page was written and reports the word counts.
The payoff is that you can mass-draft content while doing something else entirely. The same delegation pattern works for drafting emails, generating markdown files, or building dashboards, which is the same philosophy behind delegating real marketing tasks to AI on your desktop.
Why this beats doing it inside Notion alone: Notion has GPT-5.5 built in and can write a blog on command, but launching several writers at once from inside Notion is clunky and tends to work one at a time. Sub-agents in Hermes spin up multiple writers in parallel and let you monitor each session live.
Generating ChatGPT Images 2.0 Inside Hermes
Yes. With Codex selected as your model, you can generate images at full ChatGPT Images 2.0 quality without opening ChatGPT at all. You describe the image in a prompt, Codex refines it, and it renders inside Hermes using your subscription.
In the demo, the prompt was a World Cup 2026 scene: a soccer field with the U.S. flag draped over it, a flyover of helicopters, players on the pitch, and a stadium POV. Codex generated it on the spot, and then edited it on request, repositioning a goal and removing goalposts, the same conversational editing you would expect inside ChatGPT. For the bigger context on this release, see the breakdown of ChatGPT Images 2.0 and workspace agents for marketers.
The practical win for marketers: while your sub-agents are writing blogs, you can be generating cover images for those same posts in the same window, then tie the whole package together before anything is published.
Delegating Coding Tasks to Codex
You can hand coding work to Codex in two ways: launch Codex directly as the runtime, or tell any model to delegate the build to Codex. That means you can use a cheaper model like DeepSeek V4 Flash for planning and orchestration, then say “delegate to Codex” for the actual build, or use GPT-5.5 as the orchestrator and pass the heavy lifting to Codex.
In the demo, a simple prompt produced a working to-do app. Codex planned it out, created a project folder and repo, ran the Codex CLI to generate a Vite, React, and Tailwind app, installed dependencies, ran the tests, and confirmed it launched in the browser with local storage persistence and responsive styling. The result was basic by design, but it was a complete, runnable app delivered hands-off.
From experience: The flexibility here is the point. You can plan with one model to conserve Codex sessions, then delegate only the build step to Codex. Everything (the planning chat, the code, and the running app) lives in the same Hermes session you can return to later.
Analyzing Images and Screenshots with Codex
Yes. As long as Codex is your active model, you can upload an image and have it analyzed, which is useful for tutorials, walkthroughs, or any time you need a screenshot described or interpreted. In the demo, a photo of a dog in a forest was uploaded and Codex returned an accurate analysis.
You can also set Codex as your default image analysis tool by asking Hermes to configure it (“use Codex as my default image analysis tool”). After that, every uploaded image runs through Codex automatically until you hit a session or usage limit.
Smart fallback: Hermes does not ship with a built-in default image analysis model, which is exactly why Codex is handy here. To cover yourself when you hit session limits, put a few dollars into a cheap multimodal option (an older Gemini model through OpenRouter or Google’s API works well) as a low-cost backup.
Why One Hermes Workbench Beats Separate Apps
The core argument is consolidation. You are already paying for the Codex subscription, and Hermes is free to download, so routing your writing, design, development, and analysis through one app removes the constant context-switching between ChatGPT, a coding tool, and your notes app. Different sessions can even run on different models at the same time, so Codex handles your build while a cheaper model drafts emails or does research.
| Task | Separate apps | One Hermes workbench |
|---|---|---|
| Writing multiple blogs | Draft one at a time across tabs | Launch parallel sub-agents in one prompt |
| Image generation | Open the ChatGPT app separately | Generate and edit ChatGPT Images 2.0 in-window |
| Coding | Switch to a dedicated coding tool | Delegate the build to Codex in the same session |
| Image analysis | No native default; jump tools | Upload and analyze with Codex inline |
| Model flexibility | Locked into each app’s model | Run different models per session at once |
| Cost | Multiple subscriptions and tools | Reuse one Codex plan inside a free app |
In one roughly half-hour stretch, the workflow produced three published blogs, several generated and edited images, an image analysis, and a working app, all from one desktop app. If you want help building this kind of AI-powered workflow into your own marketing operation, take a look at the marketing services at Jason Pollak Marketing.
Frequently Asked Questions
Do I need an API key to use Codex in Hermes?
No. Connecting is a one-click browser login and authorize flow inside the Providers settings. There is no key to generate or paste.
Which ChatGPT plan do I need?
Any plan that includes Codex works. Hermes uses whatever tier you are on (Plus, Pro, and so on), and higher tiers give you more session and weekly usage.
How does Codex billing work inside Hermes?
It runs on your subscription’s session and weekly limits, specifically a five-hour session limit and a weekly cap. Higher effort settings like Max and Fast use up that allowance faster.
Can I run other models at the same time as Codex?
Yes. Each session can use a different model, so you can run Codex for a build while a cheaper model like DeepSeek V4 Flash handles other tasks in parallel.
Can sub-agents do more than write blogs?
Yes. Sub-agents can be delegated to almost any task, including drafting emails, generating markdown files, building dashboards, or running coding jobs.
Does this work in the terminal too, or only the desktop app?
Both. Hermes works in the desktop app and the terminal, and they share the same sessions. The desktop app is just easier for setup and for monitoring multiple agents at once.
About Jason Pollak
Jason Pollak is a marketing strategist with over 10 years of experience building campaigns for entertainment brands, artists, and businesses across music, film, television, eCommerce, and B2B SaaS. As Director of Marketing at Young Money Entertainment, he grew Lil Wayne’s Facebook following from 10 million to 50 million and managed over 60 million followers across the roster. He also served as Paid Media Director at Horizon Media, launching major TV shows for History Channel, A&E, WWE, and Lifetime, and led film marketing for Utopia Distribution, generating over $10 million in revenue on a $200K media spend. Jason specializes in paid media, organic social strategy, email automation, SEO, content development, and AI-driven marketing systems. He holds a BA in English Literature from Binghamton University and a Masters in Media Studies from Brooklyn College. Learn more at jasonpollakmarketing.com or schedule a call to discuss your marketing.