How to Use OpenClaw with ChatGPT Plus: Complete Setup Guide (2026)

Why This Topic Matters Right Now

Keyword discovery for "openclaw" shows a clear cluster of high-opportunity searches: **openclaw vs claude code**, **openclaw with gemini**, **openclaw with chatgpt**, **openclaw with chatgpt plus**, and **openclaw free api** — all with an opportunity score of 75.

The important bit is not just the score. It's the intent. Searchers asking about **OpenClaw with ChatGPT Plus** are usually asking one of three things:

That's a real setup problem, not fluffy comparison content.

• "Can I use my existing $20/month subscription instead of API billing?"
• "Do I need an API key to get OpenClaw working?"
• "What model path should I choose in onboarding?"

The One Concept You Must Understand First

OpenAI's own help docs state that **ChatGPT and the API platform are separate platforms**, with **separate billing systems**. OpenAI also says ChatGPT Plus costs **$20/month** and that **API usage is separate and billed independently**.

That means: **ChatGPT Plus** pays for the ChatGPT app experience, **OpenAI API** pays for token-based API usage, and **OpenClaw can use either route** — but they are not interchangeable.

OpenClaw's docs make the split explicit: `openai/gpt-5.4` is the direct OpenAI Platform API route, and `openai-codex/gpt-5.4` is the ChatGPT/Codex OAuth route.

That's the whole game.

When ChatGPT Plus Is Enough — and When It Isn't

Here's the clean answer. For signing in with your ChatGPT/Codex account, use the `openai-codex/*` route — no API key needed, usually requires ChatGPT Plus or another entitled plan. For direct token-based OpenAI API billing, use the `openai/*` route — API key required, no ChatGPT Plus needed. For predictable pay-as-you-go cost control, use the `openai/*` route with an API key. To reuse your ChatGPT/Codex login inside OpenClaw, use the `openai-codex/*` route. For cloud-only OpenAI/Codex workflow features, use the `openai-codex/*` route.

If you already pay for ChatGPT Plus and want the **fastest path to a working OpenClaw setup**, start with **Codex OAuth**.

If you care more about **billing precision, automation predictability, and direct API control**, use an **OpenAI API key** instead.

For most normal users, the best answer is: **start with ChatGPT Plus sign-in, then add an API key later if you outgrow it**.

Step-by-Step: How to Set Up OpenClaw with ChatGPT Plus

**Option 1: Use ChatGPT Plus through Codex OAuth**

This is the cleanest route if you already have a ChatGPT subscription and want to test OpenClaw without building an API billing setup first.

**Step 1: Install OpenClaw.** OpenClaw recommends **Node 24** and supports **Node 22.14+**. The onboarding docs say the local gateway defaults to **port 18789**. On Windows PowerShell, use the PowerShell install script instead.

**Step 2: Start onboarding.** When prompted for provider/auth, choose the OpenAI/Codex login flow. OpenClaw also documents non-interactive and direct auth routes, but for most people the simple path is the models auth login command.

**Step 3: Set your default model.** Use the Codex route as your primary model. That tells OpenClaw to use the ChatGPT/Codex OAuth-backed route instead of the API-key-backed route.

**Step 4: Verify the gateway.** If the dashboard opens and replies, you're done.

**Option 2: Use OpenAI API billing instead**

If you want direct API access, usage-based billing, or cleaner infrastructure for team automation, use the API route.

**Step 1: Add your API key.** Use the onboarding command with the openai-api-key auth choice, or set the OPENAI_API_KEY environment variable explicitly.

**Step 2: Set the direct API model.** That uses OpenAI's direct API billing rather than ChatGPT/Codex account auth.

curl -fsSL https://openclaw.ai/install.sh | bash

iwr -useb https://openclaw.ai/install.ps1 | iex

openclaw onboard

openclaw models auth login --provider openai-codex

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "openai-codex/gpt-5.4"
      }
    }
  }
}

openclaw gateway status
openclaw dashboard

openclaw onboard --auth-choice openai-api-key

export OPENAI_API_KEY="sk-..."
openclaw onboard --openai-api-key "$OPENAI_API_KEY"

{
  "env": {
    "OPENAI_API_KEY": "sk-..."
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "openai/gpt-5.4"
      }
    }
  }
}

The Best Practical Setup: Keep Both Routes Available

This is the setup I'd recommend for most serious users. Use **Codex OAuth** as the easiest primary path if you already subscribe to ChatGPT Plus. Keep an **API-key fallback** for cases where you want direct billing or a cleaner automation path.

Why this works well: your day-to-day chat can run through your ChatGPT/Codex sign-in, your fallback stays available if the OAuth route changes, expires, or has different limits, and you can reserve API billing for higher-volume or more structured automation.

For more model selection guidance, read Best AI Model for OpenClaw in 2026: Complete Comparison Guide, OpenClaw with OpenRouter: Complete Multi-Model Access Guide, and OpenClaw + Gemini 2.5: Setup Guide & Free Tier.

{
  agents: {
    defaults: {
      model: {
        primary: "openai-codex/gpt-5.4",
        fallbacks: ["openai/gpt-5.4-mini"]
      },
      models: {
        "openai-codex/gpt-5.4": {
          alias: "ChatGPT Plus"
        },
        "openai/gpt-5.4-mini": {
          alias: "OpenAI API Fallback"
        }
      }
    }
  }
}

Cost Math: ChatGPT Plus vs OpenAI API Inside OpenClaw

This is where people get confused. ChatGPT Plus is **$20/month**. OpenAI API pricing is currently: GPT-5.4 input at $2.50/1M tokens, GPT-5.4 cached input at $0.25/1M tokens, GPT-5.4 output at $15.00/1M tokens, GPT-5.4 mini input at $0.75/1M tokens, and GPT-5.4 mini output at $4.50/1M tokens.

**Rough comparison example:** Suppose an OpenClaw workflow burns 2 million input tokens and 300,000 output tokens. At GPT-5.4 list pricing, that's roughly $5.00 for input plus $4.50 for output, totaling **about $9.50**. That means for moderate usage, the API route can still be cheaper than a monthly subscription. But once you use OpenClaw constantly across many sessions, ChatGPT Plus-backed usage may feel simpler operationally.

**What actually matters more than sticker price:** ChatGPT Plus / Codex OAuth offers high billing predictability (fixed monthly subscription), easier onboarding, but lower fine-grained cost control and weaker explicit token accounting. The OpenAI API offers variable usage-based billing, higher fine-grained cost control, better team automation fit, and better explicit token accounting.

If you care about lowering your bill, also read 5 Ways to Cut Your OpenClaw API Bill by 80% and OpenClaw Best Practices: 21 Rules for a Faster, Cheaper, More Reliable AI Agent.

Advanced Config That Is Actually Worth Using

**1) Use fast mode when you want lower latency.** OpenClaw supports fast-mode toggles for both `openai/*` and `openai-codex/*` routes. This maps to OpenAI priority processing behavior where supported.

**2) Keep transport on auto unless you have a reason not to.** OpenClaw's OpenAI docs note that `auto` tries **WebSocket first, then SSE fallback**, which is the sane default.

**3) Cap runtime context if you want tighter control.** OpenClaw documents a native context window of **1,050,000** for `openai-codex/gpt-5.4`, but a default runtime cap of **272,000 context tokens** for better latency and practical quality. You can set a lower runtime cap if your sessions are bloating and you want tighter cost and latency behavior.

{
  agents: {
    defaults: {
      models: {
        "openai-codex/gpt-5.4": {
          params: {
            fastMode: true
          }
        }
      }
    }
  }
}

{
  agents: {
    defaults: {
      models: {
        "openai-codex/gpt-5.4": {
          params: {
            transport: "auto"
          }
        }
      }
    }
  }
}

{
  models: {
    providers: {
      "openai-codex": {
        models: [
          {
            id: "gpt-5.4",
            contextTokens: 160000
          }
        ]
      }
    }
  }
}

A Clean Production-Ready Config Example

If you want one config block that makes sense for a real daily-driver setup, use something like this:

Why this is a solid default: `openai-codex/gpt-5.4` keeps the main experience tied to your ChatGPT/Codex login, `openai/gpt-5.4-mini` gives you a cheaper backup path for lightweight jobs, `transport: "auto"` lets OpenClaw use WebSocket first and fall back to SSE when needed, and `fastMode: true` is worth testing if you care more about latency than squeezing every last cent out of the route.

If you later add cron jobs, channels, or coding-heavy workloads, this layout scales cleanly instead of turning into config spaghetti.

{
  agents: {
    defaults: {
      model: {
        primary: "openai-codex/gpt-5.4",
        fallbacks: ["openai/gpt-5.4-mini"]
      },
      models: {
        "openai-codex/gpt-5.4": {
          alias: "ChatGPT Plus Primary",
          params: {
            transport: "auto",
            fastMode: true
          }
        },
        "openai/gpt-5.4-mini": {
          alias: "Cheap API Fallback"
        }
      }
    }
  },
  plugins: {
    entries: {
      openai: {
        config: {
          personality: "friendly"
        }
      }
    }
  }
}

Common Mistakes When People Try OpenClaw with ChatGPT Plus

**Mistake 1: Assuming ChatGPT Plus includes API credits.** It doesn't. OpenAI explicitly says API usage is billed separately.

**Mistake 2: Choosing the wrong provider route.** If you sign in with ChatGPT/Codex credentials but set your model to `openai/gpt-5.4`, you're mixing routes. Use `openai-codex/*` for ChatGPT/Codex sign-in and `openai/*` for API key usage.

**Mistake 3: Blaming OpenClaw for auth confusion.** A lot of "OpenClaw is broken" complaints are really just "I don't understand OpenAI's product split." That's not the same problem.

**Mistake 4: Using expensive models for everything.** You probably do **not** need GPT-5.4 for every background task, heartbeat, or low-stakes lookup. Use a smaller fallback for cheaper runs.

Troubleshooting OpenClaw with ChatGPT Plus

**Problem: Onboarding completed, but replies fail.** Run the obvious diagnostics first. OpenClaw's troubleshooting docs call this the command ladder for a reason. Start there.

**Problem: You signed in, but OpenClaw still asks for API-style auth.** Check your selected model. If your config still points to `openai/gpt-5.4`, OpenClaw will expect the API route. Fix: switch the primary model to `openai-codex/gpt-5.4`.

**Problem: ChatGPT Plus account works in ChatGPT, but behavior differs in OpenClaw.** That can happen. OpenClaw docs explicitly note that provider-side limits for `openai-codex/*` can differ from the normal ChatGPT web/app experience. Fix: don't assume identical limits, latency, or routing behavior. If you need hard predictability, add an API key fallback.

**Problem: Transport instability or streaming weirdness.** Set transport explicitly only if `auto` keeps misbehaving. Try `sse` if WebSocket negotiation is getting flaky.

**Problem: Session feels slow or too expensive.** Three fixes usually work: lower the runtime context cap, use a smaller fallback model such as `openai/gpt-5.4-mini`, or shorten bloated sessions and start fresh more often.

If your OpenClaw setup is getting heavy, read OpenClaw Troubleshooting: The Complete Guide to Fixing Common Issues and OpenClaw Memory System: How Persistent Context Actually Works.

openclaw status
openclaw gateway status
openclaw logs --follow
openclaw doctor

{
  agents: {
    defaults: {
      models: {
        "openai-codex/gpt-5.4": {
          params: {
            transport: "sse"
          }
        }
      }
    }
  }
}

Recommended Setup Patterns

**Pattern A: "I already pay for ChatGPT Plus and just want OpenClaw working."** Use `openclaw onboard`, Codex/OpenAI sign-in during onboarding, and set primary model to `openai-codex/gpt-5.4`. Best for solo users, fast setup, minimal friction.

**Pattern B: "I need cleaner cost control."** Use an OpenAI API key, set primary model to `openai/gpt-5.4-mini` with `openai/gpt-5.4` as fallback. Best for cost-sensitive users and heavier automation.

**Pattern C: "I want both."** Use ChatGPT Plus / Codex OAuth as primary, OpenAI API as fallback, and a lower-cost mini model for background tasks. Best for power users who want convenience without losing control.

Conclusion

If you're searching for **how to use OpenClaw with ChatGPT Plus**, the answer is refreshingly simple once you strip away the marketing fog: **Yes, OpenClaw works with ChatGPT Plus. No, ChatGPT Plus does not replace API billing.** Use `openai-codex/*` for ChatGPT/Codex sign-in, and use `openai/*` for API key billing. If you want the fastest path, use Codex OAuth first. If you want stricter cost control, add the API route second. That hybrid setup is the sweet spot. And if someone tells you "just use ChatGPT Plus as an API key," they have no idea what they're talking about.

Frequently asked questions