Onyx CLI 1.4.0 turns its gateway into a drop-in AI endpoint

Onyx released CLI version 1.4.0 on August 24, and the headline change is compatibility, not features. The gateway adds native Anthropic passthrough for /v1/messages and serves the OpenAI Responses API natively for true OpenAI models. SDK clients built for either provider can point straight at Onyx with no shim translating requests in between, according to the release notes.

That matters for anyone running a mixed fleet of models behind one gateway. Translation layers are where bugs live: a header gets dropped, a streaming format breaks, a token count comes back wrong. Removing the shim removes a whole category of failure mode, and teams don't have to fork or patch their existing Anthropic or OpenAI client code to route through Onyx.

The release also adds a /model switcher with the current model shown in the status bar, so a session's target is visible instead of buried in config. On the auth side, JWT support takes optional audience and issuer enforcement, tightening what a token has to match before Onyx accepts it. There's a full OAuth credential lifecycle for Salesforce connections, and a fix for duplicate Google Drive group-sync crawls that were stacking up and running the process out of memory.

None of this is a new product surface. It's a gateway getting less opinionated about how you talk to it, and that's exactly the kind of change that decides whether a team keeps a custom proxy or rips it out.