This week, Anthropic, the AI research lab, made a significant announcement: the reintroduction of Claude CLI within its cloud-based AI tool, OpenClaw. This change allows developers to utilize the Claude model with greater flexibility, integrating existing API keys and token profiles. As a result, AI developers can expect a more streamlined experience when working with Claude, enhancing their productivity and convenience.

Claude CLI Reintroduction and API Key Integration

According to Anthropic's announcement, the use of Claude CLI is now permitted in OpenClaw, and existing Anthropic API keys and token profiles remain valid. Users can authenticate non-interactively by selecting their Anthropic API key in the OpenClaw settings. For instance, developers can set their API key using the following command:

bash
openclaw onboard --anthropic-api-key "$ANTHROPIC_API_KEY"

Additionally, existing Anthropic token profiles will continue to be recognized at runtime, allowing developers to maintain their previous environments without any new configurations.

New Features and Configuration Options

The Claude 4.6 model incorporates adaptive thinking by default, enabling users to redefine this behavior on a per-message basis using the `/think:<level>` command. OpenClaw also introduces a `/fast` toggle, allowing users to adjust the service layer for API requests. This feature applies only to requests directed to api.anthropic.com and does not affect those routed through proxies or gateways.

Moreover, OpenClaw supports prompt caching, which can be configured exclusively for API use. Users can set the cache retention period through the `cacheRetention` parameter, with the default set to short. For example, to extend the cache duration, developers can use the following configuration:

bash
params.cacheRetention: "long"

Anthropic also supports a 1M context window feature, which enables processing of longer contexts. To activate this capability, users need to set the following:

bash
params.context1m: true

This feature is available for specific models, allowing developers to handle more extensive context in their applications.

Conclusion

The reintroduction of Claude CLI in Anthropic's OpenClaw is poised to offer developers increased flexibility and convenience. This shift not only enhances the usability of AI models but also fosters a more efficient working environment for developers.