Step-by-step guide for OpenClaw users switching to OpenAI Codex after Anthropic and Google ToS updates. Backup, configure, and optimize for stability and code prowess.
Imagine waking up to a notification that your AI agents, humming along on Claude Pro via OpenClaw, might violate updated Terms of Service. In early 2026, Anthropic (Claude) and Google (Gemini) tightened their OAuth policies for third-party tools like OpenClaw, sparking widespread uncertainty among users relying on subscription-based access. Developers and teams using OpenClaw for autonomous agents suddenly faced the question: Is my setup still compliant?
This guide cuts through the confusion with a practical, 15-minute migration to OpenAI Codex, backed by community-tested steps. We'll cover backups, seamless switching, per-agent tweaks, and a resilient multi-layer strategy to future-proof your workflow. Whether you're optimizing code generation or maintaining productivity, you'll emerge with a stable, ToS-compliant setup leveraging Codex's strengths in tools and reasoning.
The trigger? Anthropic and Google's ToS updates explicitly question subscription OAuth in tools like OpenClaw, while OpenAI's ChatGPT Plus/Pro explicitly allows third-party Codex integration via OAuth. No more gray areas, migrate confidently.
Before diving in, assess these realities:
Prompt compatibility: Claude and GPT Codex interpret prompts differently. Remember tuning SOUL.md and AGENTS.md for optimal output.
Model strengths: Codex shines in code generation and tool use; Claude edges out on natural prose like essays or posts.
Claude alternative: Stick with API keys from console.anthropic.com, fully legal, but Opus loops can hit $100+/month.
Your OpenClaw config (channels, memory, crons) stays intact; you're just swapping the AI brain.
Protect everything first:
cd ~
tar -czf openclaw-backup.tgz .openclawThis captures channels, memory, crons, and creds in one go.
openclaw onboard --auth-choice openai-codexKey choices:
Screen | Choice | Why |
|---|---|---|
"This is risky" warning | Proceed | Flags behavioral shifts only |
Setup mode | Quickstart | Defaults for speed |
Existing values | Use existing values | Avoids wiping configs |
Log in via browser with ChatGPT account, authorize, then close the wizard immediately to prevent overwrites.
openclaw models set openai-codex/gpt-5.3-codexopenclaw models status --plainConfirm primary: openai-codex/gpt-5.3-codex. Test via your channel (e.g., Discord).
openclaw models fallbacks clear
openclaw models auth order clear --provider anthropic
# Edit ~/.openclaw/.env: delete ANTHROPIC_API_KEY
openclaw gateway restart
openclaw doctorGlobal models set doesn't touch agent-specific settings. Check openclaw.json:
{
"agents": {
"list": [
{
"id": "branding",
"model": { "primary": "anthropic/claude-opus-4-6" } // Still old!
}
]
}
}Fix via:
Cron override: openclaw cron edit <cron-id> --model "openai-codex/gpt-5.3-codex"
Delete model block: Falls back to global.
Common observations from users who’ve made the switch:
Prompt tuning is unavoidable: Budget a few days to adjust SOUL.md and AGENTS.md until things stabilize
Code and tool use are noticeably stronger: Codex delivers measurably better results for code generation and tool calls
Natural writing takes a step back: Essays and social media posts feel less polished compared to Claude
Stability is a real advantage: OAuth auth is reliable, and you can run agent loops without tracking token burn
Our team of experts will respond promptly to understand your needs and recommend the best solution.
Got questions? We've got answers.
Leonard
Co-founder
