AI provider auth must work before Discord replies will work reliably. If you are setting this up from Cue web, this is the simplest flow:
  1. In the chat page, click the workspace icon in the top right to open the workspace.
  2. In the workspace, open the top-right three-dots menu and click Open VS Code.
  3. In VS Code, create or upload:
~/.openclaw/agents/main/agent/auth-profiles.json
Use the example auth-profiles.json further below on this page as the starting format.
  1. Open a terminal in VS Code with `Ctrl+“ or from the top-left menu via Terminal.
  2. Point OpenClaw at that profile explicitly:
openclaw config set auth.order.openai-codex '["openai-codex:default"]' --json
openclaw config set agents.defaults.model.primary openai-codex/gpt-5.4
openclaw config set agents.defaults.model.fallbacks '[]' --json
  1. Verify the runtime sees the profile:
openclaw models status --json
This workflow is useful when you already have a working auth-profiles.json and want to reuse it inside the sandbox without going through interactive login first.

​
Anthropic API Key

If you want to use Anthropic in the sandbox: 
openclaw config set auth.profiles.anthropic:default '{"provider":"anthropic","mode":"api_key"}' --json
openclaw config set auth.order.anthropic '["anthropic:default"]' --json
openclaw config set agents.defaults.model.primary "anthropic/claude-opus-4-6"
openclaw config set agents.defaults.model.fallbacks '[]' --json
Then add the Anthropic key with the auth flow you prefer. Example ~/.openclaw/agents/main/agent/auth-profiles.json: 
{
  "version": 1,
  "profiles": {
    "anthropic:default": {
      "type": "api_key",
      "provider": "anthropic",
      "key": "sk-ant-..."
    }
  }
}

​
Set the Default Model Explicitly

If you use an API key flow, set the default model explicitly: 
openclaw config set agents.defaults.model.primary "anthropic/claude-opus-4-6"
openclaw config set agents.defaults.model.fallbacks '[]' --json
If you use OpenAI Codex OAuth, set an OpenAI Codex model instead: 
openclaw config set agents.defaults.model.primary "openai-codex/gpt-5.3-codex"
openclaw config set agents.defaults.model.fallbacks '[]' --json

​
OpenAI Codex OAuth

OpenClaw supports interactive Codex login: 
openclaw models auth login --provider openai-codex --set-default
Use this inside the same sandbox where OpenClaw will run.
Do not assume cached OAuth tokens copied from another tool or machine will refresh cleanly inside a new sandbox. If refresh fails, re-authenticate inside that sandbox.

​
OpenAI Codex Manual Config

To configure OpenAI Codex via config commands (without interactive login): 
openclaw config set auth.profiles.openai-codex:default '{"provider":"openai-codex","mode":"oauth"}' --json
openclaw config set auth.order.openai-codex '["openai-codex:default"]' --json
openclaw config set agents.defaults.model.primary openai-codex/gpt-5.4
openclaw config set agents.defaults.model.fallbacks '[]' --json
Profile fields (provider and mode) must be set together as a JSON object. Setting them individually will fail validation because the schema requires both fields.

​
OpenAI Codex in a Sandbox

If you already have a working Codex OAuth profile and want to reuse it in a sandbox, copy the real auth store into: 
~/.openclaw/agents/main/agent/auth-profiles.json
Example: 
{
  "version": 1,
  "profiles": {
    "openai-codex:default": {
      "type": "oauth",
      "provider": "openai-codex",
      "access": "<ACCESS_TOKEN>",
      "refresh": "<REFRESH_TOKEN>"
    }
  }
}
Then point OpenClaw at that profile explicitly: 
openclaw config set auth.order.openai-codex '["openai-codex:default"]' --json
openclaw config set agents.defaults.model.primary openai-codex/gpt-5.4
openclaw config set agents.defaults.model.fallbacks '[]' --json
The copied auth-profiles.json already carries the provider and auth type. For a seeded sandbox profile, the required follow-up is selecting it via auth.order and setting the default model. Verify the runtime sees the profile: 
openclaw models status --json
What you want:
  • defaultModel is openai-codex/gpt-5.4
  • providersWithOAuth includes openai-codex
  • missingProvidersInUse does not include anthropic

​
Final Verification

Before testing Discord, verify the assistant itself can answer: 
openclaw agent --agent main --message "Reply with OK only"
Expected: 
OK
If this command fails, fix model auth before continuing.

​
Common Failure

If you see: 
No API key found for provider "anthropic"
then the default model is still pointing at Anthropic, or no default model was set at all.