Runtimes: OpenClaw, Hermes, and Codex.

Templates decide first

Every template declares which runtimes it supports. When it supports one, the deploy form locks it; when it supports several, you pick. Choose a template for what the agent should do and let the runtime follow; the managed environment underneath is the same either way.

Hermes

Hermes is the Qoren-native runtime: agents are defined from configuration, persona, tasks, tools, workspace files, with cron, messaging gateways, and workspace support built in.

• Supports imported skills, so an agent's capabilities can be extended from a library.
• Supports approval mode: actions can wait for a human sign-off.
• Supports warm chat: conversational sessions with the running agent.

OpenClaw

OpenClaw is the right runtime when the template or workflow is built for the OpenClaw operating model: existing OpenClaw agents, skills, and configuration move to a managed environment while keeping their behavior. Qoren keeps the agent online, isolated, and metered; the OpenClaw runtime works the way OpenClaw users expect.

Codex

Codex is the runtime for code-first agents built on the OpenAI Codex CLI. It runs in the same managed environments as the other two, with the same schedules, triggers, logs, and spend controls, and appears as a runtime choice on the templates that declare it.

What is identical on Qoren

The runtime changes the harness, not the operating model. Every runtime gets the same platform underneath.

• A managed, isolated environment, each agent under its own system user.
• Vault secrets, template env vars, and managed or BYOK model keys.
• Scheduled tasks, jobs with live progress, logs, and conversations.
• Spend metering per agent and task, with the same credit balance and hard-stop behavior.