Claude Desktop becomes a versatile agent shell: third-party inference is now online, and the community has successfully run GPT and Grok.

ME News, April 23 (UTC+8): According to Beating Monitoring, Anthropic has rolled out a Third-Party Inference feature in Claude Desktop. Inference requests from the Cowork and Code tabs can be forwarded to Amazon Bedrock, Google Vertex AI, Microsoft Foundry, or a custom Gateway. Official documentation describes this feature as an enterprise deployment solution for compliance requirements: regulated enterprises that cannot send data to Anthropic, international organizations with data residency requirements, and public sector entities operating under the FedRAMP or ITAR frameworks. Configuration is pushed via MDM tools; conversation data is sent only to the inference endpoint and local machine specified by the user, and does not pass through Anthropic servers. This feature does not require an Anthropic account—simply open the configuration panel via the top menu Help > Troubleshooting > Enable Developer Mode. Some users connect third-party gateways such as OpenRouter via the Gateway interface, set the Gateway base URL to openrouter.ai/api, and call non-Claude models such as GPT-5.4 and Grok 4.20 in the Cowork interface. This is an extended community use of the Gateway interface, not a feature designed by Anthropic for non-Claude models. Hands-on testing shows that in 3P mode, only the Cowork and Code tabs are available; the Chat tab, Skills, and Plugins store cannot be used in this mode, and network egress allowlisting must be configured manually. (Source: BlockBeats)