What is BYOK?
BYOK stands for Bring Your Own Key: every plan uses an AI you already have. That can be an AI subscription you already pay for, or an API key from an AI provider (such as Anthropic, OpenAI, or OpenRouter). You pay the AI side directly, and you pay OneClickClaw separately for the hosting infrastructure. There is no built-in free model.
- On OpenClaw - Sign in with your ChatGPT plan (Plus or higher, through Codex login (device code)) or your SuperGrok or X Premium+ plan (through xAI OAuth), or paste an API key. All of it happens on your agent's Model Setup page: press Connect provider.
- On Hermes Agent - Sign in with a ChatGPT plan, SuperGrok or X Premium+, Nous Portal or a MiniMax plan, or use an API key. Choose it on the build page or from the AI Model tab of your server page.
This is fundamentally different from platforms that bundle AI usage into their subscription price. With BYOK, your AI costs are transparent, there is no markup, and you have full control over which provider and model your agent uses. If a cheaper or better model launches tomorrow, you can switch to it yourself without waiting for anyone.
Note
Your key and your sign-in live on your dedicated server, and your agent's conversations flow directly from your server to your AI provider. On Hermes Agent, a key you type into the AI Model card passes through our backend on its way to your server. We look inside servers for one purpose only: an automatic check for crypto mining. At least once an hour, and more often while a server is new or busy, it reads the list of running programs, the program files behind the busiest ones, their network connections, a few system settings and files named like known miners; on Hermes servers it also reads the commands the agent ran. It never reads your chats or documents.
Why we chose BYOK
The BYOK model exists for three important reasons:
- You control your AI costs. Token pricing varies dramatically between providers and models. A lightweight model like Claude Haiku might cost fractions of a cent per message, while a frontier model like Claude Opus costs more. With BYOK, you choose what fits your budget and can switch at any time. We do not mark up AI costs by a single cent.
- You choose your provider. You are not locked into a single AI vendor. Your agent works with 25+ AI providers. If a new model launches tomorrow that better suits your needs, you can switch to it inside your agent without waiting for us to add support.
- Your conversations stay off our systems. Your agent talks to the AI provider straight from your server, so its conversations never pass through OneClickClaw's systems.
Tip
If you are not sure which provider to use, OpenRouter gives you access to hundreds of models from dozens of providers with a single API key. It is the easiest way to experiment with different models without creating accounts at multiple providers.
Supported providers
Your agent works with 25+ AI providers. Subscription sign-ins (ChatGPT, SuperGrok or X Premium+, and on Hermes Agent also Nous Portal and MiniMax) are covered above. For API keys, here are the most popular choices:
- Anthropic (Claude) - The Claude family (Haiku, Sonnet, Opus), with an Anthropic API key. Known for strong reasoning, safety, nuanced conversations and long context windows.
- OpenAI (GPT) - The GPT family, with an API key or a ChatGPT plan sign-in. Broad general-purpose capability and fast response times.
- Google (Gemini) - The Gemini family, including the Flash models. Strong multimodal capabilities and competitive pricing, especially for the Flash models.
- xAI (Grok) - Grok models, with an API key or a SuperGrok or X Premium+ sign-in. Known for fast responses and a distinctive conversational style.
- Mistral - European AI provider offering efficient, high-quality models. Competitive pricing and strong multilingual support.
- DeepSeek - Cost-effective models with strong reasoning capabilities, especially popular for technical and analytical use cases.
- OpenRouter (hundreds of models) - A multi-model gateway that gives you access to hundreds of models from dozens of providers through a single API key. The easiest way to experiment.
Beyond these, there are more providers. On OpenClaw, the full list appears when you press Connect provider on your agent's Model Setup page.
Provider comparison
Here is a quick comparison of the most popular providers to help you choose:
| Provider | Best For | Pricing Model |
|---|---|---|
| Anthropic (Claude) | Nuanced conversations, safety, long context | Per-token (input + output) |
| OpenAI (GPT) | General-purpose, broad capability | Per-token, or your ChatGPT plan (Plus or higher) |
| Google (Gemini) | Multimodal, competitive pricing | Per-token (input + output) |
| xAI (Grok) | Fast responses, real-time knowledge | Per-token, or your SuperGrok or X Premium+ plan |
| Mistral | Efficiency, multilingual, European hosting | Per-token (input + output) |
| DeepSeek | Cost-effective reasoning, technical tasks | Per-token (input + output) |
| OpenRouter | Model experimentation, hundreds of models | Per-token (varies by model) |
How API costs work
Your OneClickClaw subscription covers hosting infrastructure only. AI usage costs are billed separately by your chosen provider. This means you have two separate, transparent bills:
- OneClickClaw: €14.99 to 49.99/month (depending on your plan) for server hosting, management, monitoring, updates, and support.
- AI provider: Billed by the provider based on token usage. Costs depend on which model you use and how much your agent is used. Typical light usage costs a few dollars per month. Heavy usage with frontier models can be more.
Tip
If you sign in with an AI subscription you already pay for, there is no separate per-token bill from that provider. With an API key, you can monitor your usage in your provider's dashboard (for example, console.anthropic.com for Anthropic or platform.openai.com for OpenAI). Setting up billing alerts on your provider account is a good way to avoid surprises.
Getting an API key
Each provider has its own dashboard where you create an API key. Here is where to go for the most popular providers:
- Anthropic: Visit console.anthropic.com, create an account, add a payment method, and generate an API key in the Settings section. Copy the key immediately (it is only shown once).
- OpenAI: Visit platform.openai.com, sign up, add billing credits, and generate an API key under the API Keys section.
- OpenRouter: Visit openrouter.ai, create an account, add credits, and generate a key from your dashboard. One key works for all of its models.
- xAI (Grok): Visit console.x.ai, create an account, and generate an API key.
- Google (Gemini): Visit aistudio.google.com, sign in with your Google account, and create an API key.
Detailed provider-specific setup guides with screenshots are available in the Setting Up Your AI Provider article.
Warning
Never share your API key in public chats, forums, or code repositories. Anyone with your key can make API calls that you pay for. If you suspect a key has been compromised, revoke it immediately from your provider's dashboard and generate a new one.
Choosing the right model
With so many models available, choosing the right one can feel overwhelming. Here is a simple framework to help you decide:
- If cost is your priority: Start with a smaller, efficient model, such as the Haiku, mini or Flash models of each family. These are fast, cheap, and handle most conversational use cases well.
- If quality is your priority: Use the frontier model of your provider. These deliver the best reasoning, instruction following, and nuanced responses.
- If you want to experiment: Use OpenRouter. With a single API key, you can try hundreds of models and switch between them inside your agent without changing your key.
You can change your model at any time: on OpenClaw under Settings, then Models, inside your agent; on Hermes Agent from the AI Model tab of your server page. For recommendations tailored to specific use cases, see Choosing the Right Model.