From the source
Your ChatGPT subscription can now power an OpenClaw agent that feels much closer to the model it is built on.
OpenClaw already supported OpenAI models, but the old path made OpenClaw drive the model loop itself.
That worked, but it also meant OpenClaw was translating between its own harness and the runtime OpenAI is actively building for agentic work.
That changes with the Codex app-server harness becoming the default runtime for OpenAI agent turns.
The short version: openai/gpt-* agent turns now run through the native Codex app-server path by default.
OpenClaw still owns the parts that make the assistant yours: channels, persona, memory, sessions, cron, media, browser, gateway, and OpenClaw tools.
Codex owns the low-level OpenAI loop: native thread state, native tool continuation, compaction, code mode, and dynamic tool search.
The lessons from this work are already flowing back into the default OpenClaw harness, so every model gets better over time.
To start with the guided path: openclaw onboard Or sign in directly: openclaw models auth login --provider openai Why the runtime matters OpenClaw is not just a chat UI.
…
