OpenClaw

OpenClaw is an open-source agent. You can add HiAPI as a model provider with its built-in setup wizard or with commands, then chat in the Dashboard (the console in your browser) or the TUI (the chat interface in your terminal). The steps below use claude-opus-5-5 as the example; HiAPI uses the OpenAI Chat Completions protocol. This tutorial was verified on OpenClaw 2026.10.7.

1. Install OpenClaw

The official install script is recommended; it checks for and prepares the Node.js environment it needs. On Mac / Linux / WSL2, run:

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

On Windows, use PowerShell:

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

After installing, reopen the terminal and run openclaw --version; seeing the version number is enough.

2. Initialize and open OpenClaw

First prepare a HiAPI API key. To use another model, find its name in the text model list and enter it.

On first use, run this in a terminal:

openclaw onboard --install-daemon

Complete the basic setup as prompted. The current version opens the model settings page in your browser; if the classic terminal wizard appears, choose QuickStart, keep main as the default agent, choose Skip for now for the model, channels, search and optional Skills, then connect HiAPI as described below. If you have already initialized, open the Dashboard directly:

openclaw dashboard

OpenClaw runs on a background Gateway service. If the Gateway is not running yet, run openclaw gateway run in one terminal and keep the window open, then run openclaw dashboard in another terminal. If you installed the background service, you can also restart it with openclaw gateway restart.

The screenshots below come from a standalone demo instance of OpenClaw 2026.10.7 on October 28, 2026; the port may differ from your machine, so use the address that openclaw dashboard opens.

3. Add HiAPI in OpenClaw

The first time you enter the Model settings page, find Custom OpenAI/Anthropic-compatible endpoint and click Set up endpoint. If you have already initialized, go to Settings → Models → Connect provider → On this Gateway to reach the same entry.

3.1 Enter the API address

Set API Base URL to https://api.hiapi.ai/v1, without adding /chat/completions, and click Submit.

3.2 Enter the API key

At the How do you want to provide this API key? step, choose Paste API key now, paste your HiAPI API key, and click Submit.

3.3 Choose the protocol

For Endpoint compatibility, choose OpenAI-compatible — Uses /chat/completions (that is, the OpenAI Chat Completions protocol). Use this protocol even when the model is Claude.

3.4 Enter the model and provider name

Set Model ID to claude-opus-5-5 and click Submit.

Set Endpoint ID to hiapi and click Submit. The full model reference in OpenClaw is then hiapi/claude-opus-5-5.

Model alias (optional) is an optional alias; you can leave it empty and submit. When verification finishes, the first-time setup shows “Inference is ready” and “Verified model: hiapi/claude-opus-5-5”.

4. Set the default model and verify a chat

Set the default model for new sessions, then check the configuration:

openclaw config set agents.defaults.model.primary hiapi/claude-opus-5-5
openclaw config validate
openclaw config get agents.defaults.model.primary

The result should include Config valid, and the last command should print hiapi/claude-opus-5-5, which means the configuration was written. You still need to verify with an actual message:

1. If you are still on the first-time setup page, click Continue with the web app, then Back to app.
2. Start a new chat, click the model name below the input box, and choose the model you just configured under the Hiapi group. Existing sessions may still use the model chosen earlier.
3. Send “Hello, introduce yourself in one sentence.” and wait for a reply.

A normal reply means the connection works. Replies may differ each time; go by the currently selected provider, whether the request succeeds, and the Request logs in the HiAPI console.

To chat in the terminal, run this while the Gateway is running:

openclaw tui

5. Write the configuration with commands (optional)

You can also do everything with commands: initialize → write the configuration → verify. Skip this section if you already configured it in the web UI above. First replace YOUR_HIAPI_API_KEY in the commands with your HiAPI API key, then run the command for your system.

openclaw config patch --stdin <<'JSON'
{
  "agents": {
    "defaults": {
      "model": {
        "primary": "hiapi/claude-opus-5-5"
      }
    }
  },
  "models": {
    "mode": "merge",
    "providers": {
      "hiapi": {
        "baseUrl": "https://api.hiapi.ai/v1",
        "apiKey": "YOUR_HIAPI_API_KEY",
        "api": "openai-completions",
        "models": [
          {
            "id": "claude-opus-5-5",
            "name": "Claude Opus 5.5"
          }
        ]
      }
    }
  }
}
JSON

@'
{
  "agents": {
    "defaults": {
      "model": {
        "primary": "hiapi/claude-opus-5-5"
      }
    }
  },
  "models": {
    "mode": "merge",
    "providers": {
      "hiapi": {
        "baseUrl": "https://api.hiapi.ai/v1",
        "apiKey": "YOUR_HIAPI_API_KEY",
        "api": "openai-completions",
        "models": [
          {
            "id": "claude-opus-5-5",
            "name": "Claude Opus 5.5"
          }
        ]
      }
    }
  }
}
'@ | openclaw config patch --stdin

This patch adds the hiapi provider and sets the default model; other providers’ configuration is kept. If a hiapi provider already exists, the patch updates its configuration and replaces the models array, so write all existing models into the array first if it had several.

On Windows CMD, save the JSON above as hiapi.patch.json, then run:

openclaw config patch --file hiapi.patch.json

After writing, check the configuration as in step 4 and verify with a new chat. If it says the Gateway needs a restart: run openclaw gateway restart if you installed the background service; if it runs in the foreground, press Ctrl+C in that window and run openclaw gateway run again.

6. Troubleshooting

Only issues specific to OpenClaw are listed here. For common errors such as 401, 404 and 429, see Common errors.