Is my data secure and encrypted?

All data is encrypted at rest using AES-256 and in transit using TLS 1.3. Your encryption keys are stored exclusively on your dedicated VPS.

Are my conversations truly private?

Yes. Each user gets a fully isolated container environment. We follow a zero-knowledge architecture.

Do you train AI models on my data?

Never. Your conversations and data are never used to train AI models.

Switching LLM Models in OpenClaw: Configuration Patterns & Best Practices

TL;DR

OpenClaw supports dynamic LLM provider switching without agent redeployment
Use `openclaw configure` wizard for interactive changes or edit `~/.openclaw/openclaw.json` directly for automation
Authentication supports both API keys and OAuth tokens; key rotation requires config reload
Model changes take effect immediately after wizard completion or gateway restart
Validate changes by inspecting gateway logs and testing agent responses with `openclaw agent test --model`

Illustration of an AI agent architecture diagram showing multiple LLM providers connected to an OpenClaw gateway with configuration files

Model Provider Landscape for AI Agents

OpenClaw abstracts provider-specific APIs behind a unified interface, enabling rapid experimentation with different LLMs for agent reasoning, tool use, and memory management.

Provider	Best For	Cost Tier	Tool Use	Latency	Auth Method
Anthropic Claude 3.5	Complex reasoning & planning	$$$	✅ Native	<800ms	API Key / OAuth
OpenAI GPT-4	General agent tasks	$$$	✅ Native	<600ms	API Key
OpenRouter	Multi-provider failover	$-$$$	✅ Proxy	<1200ms	API Key
Moonshot AI	Chinese language agents	$$	⚠️ Limited	<900ms	API Key
XAI Grok	Real-time knowledge	$$	❌ No	<700ms	API Key

Configuration Methods Compared

Two primary approaches exist: interactive wizard for manual changes and direct config editing for infrastructure-as-code pipelines.

Method	Use Case	Speed	Error Risk	CI/CD Friendly
Interactive Wizard	Development & testing	Medium	Low	❌ No
JSON Config	Production automation	Fast	Medium	✅ Yes
Environment Variables	Container deployments	Fast	Low	✅ Yes

Interactive Configuration Wizard

Launch the wizard from your OpenClaw environment terminal:

# Connect to your OpenClaw instance terminal
openclaw configure

# When prompted, select 'Model' section
# Choose provider: e.g., 'anthropic'
# Paste API key when requested
# Select default model from the displayed list
# Save and exit to apply changes immediately

Programmatic Configuration File

For automated deployments, edit the configuration directly. Note that model changes require a gateway reload.

{
  "gateway": {
    "mode": "local",
    "port": 18789,
    "token": "${OPENCLAW_GATEWAY_TOKEN}"
  },
  "model": {
    "provider": "anthropic",
    "default_model": "claude-3-5-sonnet-20241022",
    "auth": {
      "type": "api_key",
      "key_name": "anthropic_prod_key",
      "key_value": "${ANTHROPIC_API_KEY}"
    },
    "fallback_models": [
      "claude-3-haiku-20240307"
    ],
    "retry_on_failure": true,
    "max_retries": 3
  },
  "channels": {
    "telegram": {
      "enabled": true,
      "token": "${TELEGRAM_BOT_TOKEN}"
    }
  }
}

Diagram showing configuration flow from JSON file to OpenClaw gateway, with arrows indicating authentication and model selection paths

Authentication Security Patterns

Store API keys in environment variables, never commit to version control
Use short-lived OAuth tokens where supported (Anthropic, Google)
Rotate keys quarterly; OpenClaw reloads config on SIGUSR1 signal
Restrict API keys to specific IP ranges if provider supports it

⚠️ Warning: Hardcoding API keys in configuration files exposes credentials to any process with filesystem read access. In multi-tenant environments, use vault integration or encrypted secrets. OpenClaw logs mask key values, but debug mode can leak them. Never enable debug logging in production.

Validation and Testing Strategy

After model changes, verify functionality without disrupting active conversations.

Run `openclaw gateway status` to confirm model provider is active
Execute test prompt: `openclaw agent test --prompt ’What model are you?’ --expect-model claude-3-5`
Monitor token usage metrics in OpenClaw Control UI under Diagnostics > Model Usage
Check error rates: `tail -f /var/log/openclaw/gateway.log | grep ERROR`