Runtimes on Qoren: OpenClaw, Hermes and Codex
What a runtime is, how OpenClaw, Hermes and Codex differ in practice, what stays the same on Qoren, and how to choose one when you create an agent.
On this page
The runtime is the agent software that runs inside your environment. It is what reads the agent's instructions, calls the model, uses tools and keeps its memory. Qoren runs three runtimes: OpenClaw, Hermes and Codex. Most of the time you do not have to choose, because the template you start from already picked the one it was built for.
Let the template decide#
Every template says which runtimes it supports. When you create an agent, the Runtime cards start on the template's choice:
- If the template supports one runtime, that card is selected. You can still pick another, and Qoren saves a copy of the template with that runtime added.
- If it supports several, you pick the one to use.
Choose the template for the job you want done and keep its runtime unless you have a reason to change it. The environment underneath, the billing and the console are the same whichever runtime you pick.
Hermes#
Hermes Agent is an open-source agent from Nous Research. On Qoren it is defined by configuration: its persona and instructions, scheduled tasks, tools and workspace files, with chat-app gateways and a scheduler built in. Its own documentation is at hermes-agent.nousresearch.com.
Choose Hermes when you start fresh from a Qoren template, or when you design an agent with AI: the designer writes blueprints for Hermes.
OpenClaw#
OpenClaw is an open-source agent with its own gateway and chat channel plugins. It is the right runtime when a template or workflow is built for OpenClaw, or when you already run OpenClaw agents yourself: their skills and configuration keep working the way OpenClaw users expect, and Qoren keeps them online, isolated and metered. To bring an existing agent over, see migrate an agent to Qoren.
Codex#
Codex is OpenAI's command-line coding agent. Qoren runs it through codex exec, one run per message or task, which suits code-first work such as reading a repository and writing release notes. You can also sign a Codex agent in to a ChatGPT account so it runs on your ChatGPT plan instead of the model key; see sign in to ChatGPT for Codex.
What differs in the console#
The console looks the same for every agent. A few features depend on what the runtime supports:
| Feature | Hermes | OpenClaw | Codex |
|---|---|---|---|
| Chat on the agent's Chat tab | Yes | Yes | Yes |
| Telegram, Slack or Discord through Qoren | Yes | Yes | Not yet |
| Scheduled tasks from a template | Yes | Yes | Only on accounts that run on Qoren's scheduler |
| Imported skills (Settings, Brain) | Yes | No, OpenClaw manages its own skills | No |
| A saved Brain edit reaches the agent | From its next new chat or scheduled task | From its next message | From its next new chat or scheduled task |
| Session detail on the Usage tab (tokens, sources, models) | Yes | No, charged credits only | No, charged credits only |
| ChatGPT sign in (Settings, Keys) | No | No | Yes |
Where a runtime lacks a feature, the console leaves it out or says so. For example, a template with scheduled tasks cannot be created on a runtime that cannot run them: the create form shows a note and keeps Create agent off. See add skills to an agent, identity and instructions and agent usage for the features themselves.
What stays the same#
The runtime changes the agent software, not the way Qoren runs it. Every runtime gets:
- A managed, private environment, with each agent running as its own system user.
- Keys from the vault, and either Qoren's managed model key or your own.
- The same tabs: Chat, Shell, Activity, Workspace and Settings, with backups, repair and logs.
- The same credits balance and spend controls.
Choose at a glance#
| Your situation | Runtime |
|---|---|
| The template supports only one runtime | Keep that one |
| You are starting fresh from a Qoren template | Hermes, unless the template says otherwise |
| You already run OpenClaw agents, or the workflow is built for OpenClaw | OpenClaw |
| The job is code-first, or the template is built for Codex | Codex |
| People must reach the agent from Telegram, Slack or Discord | Hermes or OpenClaw |
Frequently asked questions#
Can I switch an agent's runtime later?
No. The runtime is chosen when the agent is created, and its Settings only show it. To change it, create a new agent from the same template on the other runtime, then remove the old one.
Do all templates support every runtime?
No. Each template lists the runtimes it supports, and the create form starts on one of them. You can pick another, in which case Qoren saves a copy of the template with that runtime added and warns that results may differ from what the template recommends.
Is one runtime cheaper than another?
The runtime itself costs nothing extra. You pay for the environment by its size and for model use by the model, whichever runtime runs the agent. See how credits work.