HomeGlossaryAgent runtime

Agent runtime

Also called: agent harness, agent framework

Definition

An agent runtime is the software harness that runs an agent: it holds the loop, manages context, and exposes the tools the agent is allowed to call.

In practice

The runtime is not the model and not the host. The model supplies the reasoning, the host supplies the machine, and the runtime is the layer between them that decides how the agent is configured, what a tool call looks like, and how conversation and state are kept. OpenClaw, Hermes, and the OpenAI Codex CLI are all runtimes. Because the runtime shapes how you author an agent, moving between them is a rewrite rather than a setting. Choose the host first, because that decides whether an agent can run on a schedule at all, then choose the runtime.

Example

A team runs two agents: a research agent built on Hermes and a coding agent built on the OpenAI Codex CLI. Both run on the same managed host, with the same kind of schedules, secrets, and spend cap. What differs is the runtime: how each agent's instructions are written, how its tools are declared, and how it keeps memory between runs. Moving the research agent to OpenClaw would mean rewriting that configuration, not flipping a switch.

Common questions about agent runtime

Is an agent runtime the same as a language model?

No. The language model does the reasoning, while the agent runtime is the harness around it: it runs the loop, assembles the context for each turn, and defines which tools the agent can call. The same model can sit inside different runtimes, and each runtime will make it behave differently.

Can I switch agent runtimes later?

You can, but plan for a rewrite rather than a setting change. Each agent runtime has its own way of writing instructions, declaring tools, and storing memory, so an agent built for OpenClaw does not run unchanged on Hermes. Choose the host first, then the runtime that fits the job.

Read more

Nearby terms

All 25 terms

Run agents without the homework

Qoren runs OpenClaw, Hermes, and Codex agents in managed environments, with the restarts, secrets, schedules, and spend caps handled for you.