Pairing Hermes with Codex means using subscription capacity you already pay for, inside a system that actually remembers your projects. Here’s how to set it up as its own profile, what the combination is genuinely good for, and when it’s the wrong answer.
Short answer
- Hermes is the harness; Codex is the brain — you get memory, skills and scheduling around it.
- Give it a dedicated profile so memory and history stay separate and comparisons stay fair.
- Test a tool call, not a chat — that’s where setups actually fail.
- It’s not the free or private route; use a free or local model in the profile for that.
Why pair Hermes with Codex at all
Both are capable on their own, so the honest question is what the combination gives you that either doesn’t.
Hermes is a harness — persistent memory, skills, profiles, cron jobs, and the ability to run any model you point it at. Codex is OpenAI’s coding agent, strong at the thing it was built for and backed by a subscription many people already pay for.
Pairing them means you use capacity you’re already buying, inside a system that remembers. Codex on its own starts each piece of work without your business context. Hermes supplies that context and the scheduling around it.
| Hermes | Codex | |
|---|---|---|
| What it is | A harness — memory, skills, scheduling | A coding agent |
| Cost | Free and open source | Your existing subscription |
| Memory | Persistent, per profile | Per session |
| Model choice | Any — frontier, free, local | OpenAI |
| Runs unattended | Yes, via cron jobs | Not by design |
Set it up as its own profile
The habit that makes this work, and the one people skip.
- Create a dedicated Hermes profile for Codex rather than switching your main one over.
- Point that profile at Codex and confirm it’s authenticated before you give it real work.
- Test a tool call, not a chat. Ask it to do something — read a file, run a command — rather than asking it a question. Plenty of setups answer fine and fail the moment a tool is involved.
- Give it a job. A profile with a clear role beats a general-purpose one, because the persona shapes every response.
- Run the same task in a second profile on a different model and compare properly.
Keeping it in its own profile means separate memory, separate history and a fair comparison — and when something better ships, you add a profile rather than unpicking your setup. That’s the architecture doing the work: see how the layers fit together.
🔥 Want this set up without the guesswork? Wiring a second agent in without your setup turning into a mess is exactly the sort of thing that eats an afternoon. Inside the AI Profit Boardroom you get the Agent OS as a ready-to-install file, a 30-day roadmap, daily tutorials the same day new tools ship, and four live coaching calls a week where you share your screen and get unstuck. → Get access here
What the pairing is actually good for
- Coding work that needs your context. Codex brings the capability; Hermes memory brings the knowledge of your projects, conventions and past decisions.
- Scheduled builds and checks. Cron jobs let a coding agent do things on a timetable, which isn’t how Codex works alone.
- Using subscription capacity for agent work rather than paying twice for API access on top.
- Handing work between agents. With Bot Mode, a Codex-backed bot can be mentioned by another bot and hand its answer back in the same conversation.
The last one is the genuinely new capability. A research bot on one model, a Codex bot doing the build, and the results flowing between them without you copying and pasting between windows.
Where it isn’t the right answer
Being straight about this saves you time.
- If you want free. Codex means a subscription. Free routes exist — see the free API setup and open source models.
- If you want local and private. Codex is a cloud model. For work that can’t leave your machine, use a local model.
- If the work isn’t code-shaped. Codex is a coding agent. For research, content and general agent tasks, other models suit better.
- If you want one tool. This is a two-part setup by design. If that sounds like overhead, it probably is for you right now.
Running it alongside everything else
The setup I’d actually recommend isn’t Hermes-plus-Codex. It’s Hermes with several profiles, one of which is Codex.
| Profile | Job |
|---|---|
| Codex | Coding and build work, on subscription capacity |
| A frontier model | Planning, hard reasoning, final review |
| A cheap high-volume model | Monitoring, drafting, routine steps |
| A local model | Anything confidential |
They share one memory underneath, so what one profile learns the others can see. That’s the difference between owning a collection of agents and running a system.
Related setups: Hermes with Codex workflows, the connection guide, and the Claude Code pairing.
Want several agents working as one system? The Agent OS in the AI Profit Boardroom runs Hermes, Claude, Codex and OpenClaw side by side with shared memory, so adding one is a profile rather than a rebuild. Install file, 30-day roadmap, daily tutorials and four coaching calls a week. Start free with the free AI course and community.
FAQ
Can Hermes Agent use Codex?
Yes. Set it up as its own Hermes profile pointed at Codex, so it keeps separate memory and history from your other agents.
Why use Codex inside Hermes rather than on its own?
You get persistent memory of your projects, skills the agent keeps, scheduling via cron jobs, and the ability to hand work between agents — none of which Codex does alone.
Does it cost extra?
Hermes is free and open source. You’re using the Codex subscription capacity you already pay for rather than buying separate API access.
Should I test with a chat message?
No — test a tool call. Setups routinely answer questions fine and then fail the moment the agent has to actually do something.
Can it run unattended?
Through Hermes cron jobs, yes. That’s one of the clearest reasons to run a coding agent inside a harness.
Is Codex the right model for everything?
No. It’s a coding agent. For research, content and general agent work, point other profiles at other models.
Can a Codex bot work with my other bots?
In Bot Mode, yes — another bot can mention it, it does the work with its own memory and model, and hands the answer back in the same conversation.
What if I want this free instead?
Run a free or local model in the profile instead. The harness doesn’t care which brain you plug into it.
The bottom line
Hermes agent Codex works best as one profile among several rather than a whole setup: Codex for the build work on capacity you already pay for, a frontier model for planning, something cheap for volume, and a local model for anything confidential — all sharing one memory underneath. Give it its own profile, test a tool call rather than a chat, and let the harness do the remembering.
About Julian Goldie
I run Goldie Agency, a 7-figure SEO agency, and teach this daily on a 400K+ subscriber YouTube channel. 240+ client projects on Upwork at a 100% job-success score, 10+ years through every major Google update. My systems are in the AI Profit Boardroom; my link building book is free here.