llm 0.32

The llm 0.32 release introduces four notable additions to the command-line interface. Users can now request reasoning traces that show intermediate steps taken by the model. Integration with the OpenAI Responses API allows calls that go beyond simple completions. Server-side tool support lets external functions be invoked from within the LLM workflow.

Adopting the new version requires updating the llm package to 0.32, which may involve adjusting existing scripts that rely on the older output format. The reasoning trace feature adds extra output that parsers must be prepared to handle. Using the OpenAI Responses API necessitates a valid API key and may incur usage charges according to OpenAI’s pricing. Server-side tool support expects users to define and expose tools that the CLI can call.

The tool remains a command-line utility, so it does not provide a graphical interface or built-in web dashboard. Features that depend on model capabilities not exposed through the API will not be available. Functions that require fine-grained control over model internals beyond what reasoning traces reveal are still out of reach. Additionally, any model provider that does not implement the Responses API or tool calling cannot be used with those specific new features.

Only one source, Simon Willison’s weblog, announced this release, so there is no independent corroboration from other feeds. The lack of multiple reports means we cannot verify whether any additional changes were made beyond those highlighted. Engineers should treat the announced features as the primary source of information until further documentation appears.

The release note does not specify any breaking changes, but users should review the updated help output for new flags and options. As with any open-source project, future updates will depend on community contributions and the maintainer’s roadmap. Keeping an eye on the project’s repository will help anticipate further evolutions.