API Keys & BYOK on OpenClaw Launch

What Is BYOK?

BYOK — bring your own key — means you paste an API key from a provider you already pay (or one with a free tier) into your OpenClaw Launch account. Your running bots then call that provider directly using your key, instead of going through the credits bundled with your plan.

Why do it? Three reasons. First, cost and speed: going direct to a provider can be faster than the same model proxied through a router, and depending on the provider it can be cheaper — compare the two prices before assuming, since for some models the router is currently the cheaper route. Second, access: a handful of provider-only image and video SKUs run only when you supply that provider’s own key — though plenty of media models, including GPT Image 2 and Grok Imagine video, are already on the included catalogue without one. Third, control: usage shows up on your provider dashboard and counts against your quota, not ours.

Keys are stored encrypted in our database and, for most providers, applied to your existing running instances with no redeploy — the exceptions are called out where they apply below. The page only ever shows a saved key back to you as a masked preview. Most cards also test the key against the provider as you save, but for nearly all of them that test is advisory: the key is stored either way and the page tells you if the check did not come back clean. So treat a validation warning as real — a bad key can sit saved and quietly fail later.

One thing worth knowing before you click Save: your bot usually restarts. Switching the primary model restarts it on both OpenClaw and Hermes — the model is read at startup, so a running bot has to come back up before it uses the new one. Saving a key varies by provider: some apply live, some restart the bot, and a few never touch the container at all. Expect the bot to be away for a few seconds on OpenClaw and up to roughly 10–30 seconds on Hermes; it comes back on its own. Everything on its persistent storage — conversations, memory, workspace files and scheduled tasks — survives. Anything you installed by hand inside the container outside that storage may not. The practical advice: don’t save keys or switch models in the middle of a chat, and never while you are pairing a messaging channel.

Also worth knowing: saving a chat key sometimes does more than store it. Saving an xAI key switches your bot’s primary model to Grok straight away, without you touching the model picker — so if you want to stay on your current model, check the instance card afterwards. Anthropic and MiniMax are gentler: they only act if you are already on one of that provider’s models, and then they simply re-route it to bill your own account instead of your included credits.

Where to Find the API Keys Page

1. Sign in at openclawlaunch.com.
2. Open the API Keys page from the dashboard navigation.
3. Find the card for the provider you want, paste your key, and click Save. A green Active badge means the key is stored and switched on for your account — it is not a health check on the key itself, so still read any validation warning shown next to it.

Chat Model Providers

These keys change which model answers your bot. After saving any chat provider, use the “Set as primary model” control on the card to push that model to all of your running instances at once.

Use a ChatGPT Subscription Instead of a Key

If you already pay for ChatGPT Plus or Pro, you do not need an OpenAI API key at all. The ChatGPT Subscription card on the API Keys page connects the plan you already have through a terminal OAuth sign-in, and GPT chat then runs at no per-message API cost — nothing is drawn from your plan’s included credits and nothing is billed to a separate OpenAI account. This is a different thing from the OpenAI card above, which takes a pay-as-you-go sk- API key.

Full walkthroughs, including how to re-authenticate when the sign-in expires: ChatGPT subscription on OpenClaw and ChatGPT subscription on Hermes.

Image, Video & Music Providers

These keys power the generation tools your bot can call. After saving a media provider, pick the model you want in your instance’s image, video or music settings. Three of them — Runway, BytePlus and Replicate — currently store the key only: you can save it now, but it is not yet wired through to your instance, so nothing will generate on it. ComfyUI is different again: it is deliberately not in the model picker because it needs your own workflow and prompt-node setup rather than a ready-made one.

One Card That Is Not a Model Provider: Composio

The API Keys page also has a Composio card. It has nothing to do with which model answers your bot — Composio is what connects your bot to other apps (Gmail, Slack, Notion, Drive, GitHub and a thousand more). You do not need to add a key here to use those integrations; they work on the shared connection out of the box. Bring your own Composio key only if you already have an account and want the connections and their action limits on it instead. Browse what is available on the integrations page.

Set as Primary Model

For almost every provider, saving a chat key does not by itself switch your bot to that provider — xAI is the exception noted above. Click “Set as primary model” on the provider card and pick a model from the dropdown — this is the step people miss, and until you do it your bot keeps running on the credits included with your plan. Your running bot restarts to pick the new model up, on either framework. Media keys work differently: there is no “primary” for them; you choose the image, video or music model per instance.

You can also switch models straight from the model picker on the instance card at your dashboard. Once a key is saved, that provider gets its own section in the picker, badged “Your key” and billed to your own provider account, alongside the separate included-credits section. Where the same model exists in both, pick the one under “Your key” to keep it off your plan’s allowance — and a key often brings models the credits catalogue doesn’t carry at all. One exception worth knowing: a saved OpenRouter key takes precedence account-wide, so once you add one, OpenRouter-routed models bill to it rather than to your included credits whichever entry you pick.

What About Embeddings?

On OpenClaw, your bot uses small session-memory embeddings (Qwen3) to recall past conversation. These keep running through OpenRouter at roughly a few cents per million tokens — typically pennies a month — regardless of which chat provider you BYOK, and they can fall back to a free embedding model if the included pool runs dry. The OpenRouter card has a toggle to use your own OpenRouter key for embeddings; note that turning it off with a BYOK key in place switches cross-session memory search off rather than moving it back onto platform funding. Hermes handles its memory differently and has no equivalent embedding setting here.

Custom Endpoint

If your provider is not listed but exposes an OpenAI-compatible API (Together AI, Mistral, a self-hosted vLLM server, and so on), use the Custom API Provider card to enter the endpoint URL and key directly. See the Custom API Provider setup guide for the exact base URLs, key formats and troubleshooting steps.

OpenClaw vs Hermes Agent

The same API Keys page serves both frameworks, and nearly every key applies to whichever instances you run — classic OpenClaw bots and Hermes Agent instances alike. The exceptions are marked in the tables above: Perplexity and ComfyUI are OpenClaw-only, and Nous Portal is Hermes-native. Direct routing for Moonshot, Groq, Qwen and Z.AI is OpenClaw-side too, so on a Hermes bot those models may still be served through OpenRouter rather than your own key. Hermes instances expose image and video settings but no music picker here. The provider guides linked above generally come in an OpenClaw and a Hermes version with framework-specific notes. See the BYOK overviews: OpenClaw BYOK and Hermes Agent BYOK.

Troubleshooting

The key won’t save / shows an error

A hard save error usually means the key is malformed. Re-copy it from the provider’s dashboard, check for trailing spaces, and confirm the key prefix matches (for example sk-ant- for Anthropic, pplx- for Perplexity, gsk_ for Groq). More often, though, a wrong, revoked or unfunded key will still save and simply come back flagged as not validated — the storage step and the provider check are separate for nearly every card. If you see that flag, fix the key before relying on that provider, because the bot will fail at the first real request rather than at save time.

I saved a key but my bot still uses the old model

For almost every provider, saving stores the key without switching the model. Use “Set as primary model” on the chat provider’s card to point your instances at the new model. (xAI is the one that does switch on save.)

Image or video generation still fails after adding a media key

For most media providers the key unlocks models but you still have to select one: open your instance settings and pick a model whose provider matches the key you added (for example a fal model after adding a fal key). If the provider you added is Runway, BytePlus or Replicate, there is nothing to select yet — those keys are stored but not yet wired through to instances. ComfyUI is not in the picker by design and needs your own workflow instead, and Hermes instances have no music picker at all.

Frequently Asked Questions

Is my API key safe on OpenClaw Launch?

Yes. Keys are encrypted in our database and only ever shown back to you as a masked preview — never in plain text again. To actually call the provider on your behalf, most keys are passed into your own container, where they live in that instance’s private configuration; a few are used server-side and never reach the container at all.

Do I have to redeploy my bot after adding a key?

No — you never have to redeploy, and you never have to rebuild your bot from scratch. It does restart, though. Switching the primary model restarts the bot on both frameworks, and some key saves restart it too while others apply live. It comes back on its own — a few seconds on OpenClaw, up to roughly 10–30 seconds on Hermes — and your conversations, memory, workspace files and scheduled tasks all live on persistent storage and survive it.

Can I connect my Claude Pro or Max subscription instead of an API key?

No. Anthropic does not permit outside services to route requests through Claude Free, Pro or Max subscription credentials, so there is no supported way to wire a Claude subscription login into your bot. To run Claude on your own account, use an API key from console.anthropic.com (it starts with sk-ant-) on the Anthropic card — that is pay-as-you-go and separate from your Claude plan. A ChatGPT Plus or Pro subscription can be connected, at no per-message cost — see the section above.

Do I need a key to run Claude, GPT or Gemini at all?

No. Every paid plan already includes the same paid-model catalogue — Claude, GPT, Gemini, Grok and the rest are in the model picker on your instance card with nothing to connect, and Lite and Pro get the identical list. Without a key, that usage draws on the AI credits included with your plan, which the more expensive models use up considerably faster; when the allowance runs out, the paid models drop out of the picker until it refreshes. A key mainly changes who pays — and it usually widens the list too, since a direct provider key or a custom endpoint can offer models the included catalogue doesn’t carry.

Do I need BYOK to use OpenClaw Launch?

No. Every paid plan and the free trial include credits that work out of the box. BYOK is optional — add it when you want a specific model, lower cost, or usage on your own provider account.

Can I add more than one provider key?

Yes. You can save keys for many providers at once — for example a chat provider, an image provider, and a video provider — and each is used for its part of the pipeline.

What's Next?

• AI plans compared — subscription coding plans vs pay-as-you-go API token pricing, side by side
• OpenClaw BYOK guide — the BYOK basics for OpenClaw
• Compare AI models — pick the right model before you bring a key
• See pricing — Lite is $3 for your first month, 50% off; BYOK optional