BetterClaw
Models & Skills

Bring your own key

Power an agent with your own provider API key.

Bring your own key (BYOK)

On the Free plan, agents run on an API key you provide. BYOK is also available on every paid plan if you'd rather use your own provider account than managed credits.

Add a key

  1. Open Settings ▸ Agents and select the agent to configure.
  2. Open its Model tab.
  3. Choose your provider.
  4. Choose a model (or, for a single-model provider, the default model is shown).
  5. Paste your provider API key — the field shows the expected key prefix as a hint (for example sk-ant-... for Anthropic).
  6. Save & Redeploy.

On a paid plan the Model tab shows connected models instead: click Connect a model, pick the provider, paste the key, and choose its models. See Paid plans: connected models.

Your key is stored encrypted and used only to run that agent's model.

the Connect Anthropic step with a saved API key and the list of models to add

Replacing or changing the key

Once a key is saved, the Model tab shows it masked. Click Replace to enter a new one (on a paid plan, open the model's Edit Settings gear first). You can switch to a different model from the same provider without re-entering your key — only the model selection changes. Switching to a different provider always requires a fresh key, since a key for one provider won't authenticate against another.

OpenAI-compatible endpoints

For a self-hosted server or any endpoint not in the list, choose OpenAI Custom. Alongside the key you enter the Model ID and the endpoint's Base URL (e.g. https://api.my-provider.com/v1); the key is sent to that URL as a Bearer token. See Providers & models.

If your gateway also serves Claude over the Anthropic Messages API, choose Anthropic Custom instead to keep thinking control and prompt caching; the key is sent as x-api-key. See Providers & models.

Agents set up earlier on the legacy Custom provider keep their API key env var setting and remain editable, but Custom is no longer offered for new connections.
Treat API keys like passwords. Anyone with workspace access that can edit LLM settings can change the model and key; keep your team limited to people you trust.