Anthropic has confirmed that the Claude command‑line interface can be used inside OpenClaw once more. The company’s staff told OpenClaw developers the policy has been reinstated, meaning the bundled Claude CLI backend is again sanctioned. This change restores the workflow that many teams relied on for rapid prototyping and local testing. Teams can now script deployments without workarounds.

OpenClaw accepts both Anthropic API keys and legacy token profiles, with tokens still honored if already configured. For reliable production billing, the console‑issued API key is preferred. Models default to adaptive thinking on Claude 4.6, and the `/fast` toggle switches between `auto` and `standard_only` service tiers. Prompt caching uses a five‑minute short cache unless `cacheRetention` is overridden per agent. These defaults simplify multi‑agent configurations across environments.

Developers can also enable Anthropic’s beta‑gated 1 million‑token context window by setting `params.context1m: true` on supported Opus or Sonnet models; OpenClaw translates this to the appropriate beta header. Because legacy token auth rejects the beta request, the system falls back to the standard window and logs a warning. The feature also respects existing rate‑limit cooldowns per model, and using an API key remains the safest path.