Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.wednesdayai.dev/llms.txt

Use this file to discover all available pages before exploring further.

​
Anthropic

WednesdayAI uses Anthropic’s Claude models for its default AI. This page covers every authentication path and the Anthropic-specific config options: thinking mode and prompt caching.

​
Authentication

Obtain an API key from console.anthropic.com → API Keys. Authenticate via the interactive wizard: 
openclaw login
# choose: Anthropic
# enter your API key when prompted
Or set an environment variable before the gateway starts: 
export ANTHROPIC_API_KEY=sk-ant-api03-...
The gateway checks credentials in this order:
SourceKey / path
Environment variableANTHROPIC_API_KEY
Stored credential~/.openclaw/credentials/anthropic
Config fileproviders.anthropic.apiKey (not recommended — key ends up in config)
Never put a live API key directly in openclaw.json. Use an environment variable or the credential store so the key stays out of version control.

​
Key rotation

To rotate without downtime, pass multiple keys as a comma-separated list: 
export ANTHROPIC_API_KEY=sk-ant-new-key,sk-ant-old-key
The gateway load-balances across all keys and retries on 429 / 529 responses.

​
Setup token (Pi and embedded agents)

If you use WednesdayAI on a Raspberry Pi or headless server and cannot paste a key interactively, generate a short-lived setup token on a machine where you are already authenticated: 
openclaw setup-token --provider anthropic
On the headless device, run onboarding with the token: 
openclaw onboard --setup-token <token>
The token is exchanged for a credential and stored in ~/.openclaw/credentials/. It expires after 15 minutes.

​
Selecting a model

Set the default model in ~/.openclaw/openclaw.json: 
{
  agents: {
    defaults: {
      model: "claude-sonnet-4-6",
    }
  }
}
Available Claude model IDs:
ModelID
Claude Sonnet 4.6claude-sonnet-4-6
Claude Opus 4.7claude-opus-4-7
Claude Haiku 4.5claude-haiku-4-5-20251001

​
Thinking mode

Claude Sonnet and Opus support extended thinking — the model reasons through a problem before responding. This improves quality on complex tasks at the cost of extra tokens and latency. 
{
  agents: {
    defaults: {
      thinkingMode: "adaptive",  // "off" | "adaptive" | "always"
    }
  }
}
ValueBehaviour
offThinking disabled for all requests
adaptiveGateway decides based on query complexity (default for Claude 4.6+)
alwaysThinking enabled on every request
Users can toggle thinking per-message with /think on or /think off.

​
Prompt caching

Anthropic’s prompt caching reuses the KV cache for repeated system-prompt prefixes, cutting cost and latency on long system prompts. 
{
  providers: {
    anthropic: {
      cacheRetention: "short",  // "off" | "short" | "long"
    }
  }
}
ValueCache TTLCost savings
offNo cachingNone
short5 minutes~10% per cached request
long1 hour~25% per cached request
long is most effective when the same agent has a large, stable system prompt across many conversations.

​
Troubleshooting

The API key is invalid or has been revoked. Generate a new key at console.anthropic.com and re-run openclaw login.
Anthropic is under load. The gateway retries automatically with exponential backoff. If it persists, add a second API key to spread load — see Key rotation above.
The model ID is wrong or the model is not available on your tier. Run openclaw status --providers to list models available to your key.
Thinking requires Claude Sonnet 4.5+ or Opus 4.6+. Haiku models do not support thinking. Check your model ID in agents.defaults.model.