Connect OpenClaw
Add our API as a custom provider in an existing OpenClaw installation and choose the model your agent should use.
https://aelder.dev/v1An existing OpenClaw installation can connect directly to our API. Use its model configuration wizard; you do not need OpenCodex or a second local proxy.
Have your service API key, available balance and an exact model ID ready. Run the setup on the computer and in the OpenClaw profile that runs your agent. Copy our Base URL from the connection card above.
1. Open model configuration
Run this command in an interactive terminal:
openclaw configure --section modelThis opens the model section of the existing configuration. If OpenClaw asks which agent to configure, choose the agent you intend to use. Select Custom Provider in the provider picker.
2. Add the API connection
Use these values in the wizard:
| Prompt | What to enter |
|---|---|
| API Base URL | Our Base URL above, including /v1 |
| API Key | Your service key, without Bearer |
| Endpoint compatibility | OpenAI-compatible — the option that uses /chat/completions |
| Model ID | Select a discovered service model or enter its exact ID |
| Endpoint ID | my-api, or another recognizable unused identifier |
| Model alias (optional) | Leave blank, or choose a short label |
| Image input, if asked | Choose No for this text-only setup |
A vendor-login option is a different connection. Use the custom provider and our service key for this setup.
OpenClaw tries to discover models from /v1/models, and then makes a short generation request to verify the chosen model and format. That verification uses your API balance like a normal request. If discovery is unavailable, copy an ID from our models endpoint; do not guess a display name.
Finish the wizard and note the provider/model reference it prints, such as my-api/YOUR_MODEL_ID. If the endpoint identifier was already taken, OpenClaw can choose a different one; use the actual reference it saved.
3. Check the model selection
List the configured models and inspect the active defaults and credentials:
openclaw models list --provider my-api
openclaw models statusReplace my-api if you chose a different endpoint ID. With several agents, inspect the same agent you configured:
openclaw models list --agent YOUR_AGENT_ID --provider my-api
openclaw models status --agent YOUR_AGENT_IDAdding a provider does not necessarily replace an existing default model. In an OpenClaw chat, choose the saved model explicitly:
/model my-api/YOUR_MODEL_ID
/model statusReplace the sample model and agent IDs with your actual values. The my-api/ part is OpenClaw’s local provider identifier; the service receives the underlying model ID.
Make it the global default
For agents that inherit the global default, you can set it with:
openclaw models set "my-api/YOUR_MODEL_ID"Use the full reference from your model list. This command changes global agent defaults; it does not accept --agent and does not override an explicit per-agent model or a conversation’s selection. Configure the intended agent or use /model in that conversation when it has its own override.
4. Apply and test
Start a fresh conversation with the selected model and ask for a short reply. /model status shows the current conversation’s model; openclaw models status describes configured defaults and auth, rather than proving a live generation succeeded. Check the service dashboard for the request and actual billed usage.
If your background gateway has not picked up the saved settings, finish active work and run:
openclaw gateway restartIf you launch the gateway manually, restart it using your usual command instead of installing another service. Existing conversations can keep their own model selection after a restart, so check /model status in the conversation you are testing.
Choose another model later
Use /model in a chat to choose another registered model. For a model that is not registered yet, rerun openclaw configure --section model, choose Custom Provider again and use the same endpoint details and Endpoint ID. Register the new exact service model and check the resulting reference.
The service may expose more models than OpenClaw’s local picker. A model must be registered in the custom provider configuration and permitted by the agent’s model policy to appear there. A known name or a saved alias alone does not register a new runtime model.
If the connection fails
- Wizard cannot verify the endpoint: check the public Base URL, key, balance and selected model. Keep
/v1in the Base URL and choose the compatibility option that uses Chat Completions. - No model or provider is listed: check the profile and agent you configured. Use
--agenton list/status to inspect the correct agent. - The previous model still answers: check
/model statusin that chat. An existing agent or conversation override can take precedence over the global default. - Model is blocked or missing from the picker: check the agent’s model allowlist/policy and the custom provider’s registered models, as well as your service key’s permissions.
- Text works but an agent action fails: confirm tool-calling support and the tool’s own configuration.
This guide follows the custom-provider wizard in OpenClaw’s reviewed 2026.10.4 source. See the project’s configuration guide and model commands for version-specific details, and our API troubleshooting for service errors.
Need the key? Open your AELDER dashboard. For direct API calls, see the API reference.

