Simon Willison has released version 0.33 of his llm command-line tool, primarily to upgrade the underlying OpenAI Python library and switch the HTTP client dependency from httpx to httpx2. This update also addresses a long-standing issue where embedding models could not accept API keys on a per-call basis. The new code allows users to pass a –key flag directly to llm embed and llm embed-multi commands. This change aligns the key handling for embedding models with the pattern already used by standard language models, ensuring that plugins reading self.key continue to work without modification.
The release introduces a template chaining feature that lets users combine multiple configuration files in a single run. Developers can now save a model with default options as one template and a specific prompt as another, then execute both together to apply the settings to the text. Additionally, models supporting the Responses API can now generate a reasoning_summary with auto, concise, or detailed values. This functionality helps users test different reasoning capabilities across various OpenAI endpoint clones.
* Upgrades the HTTP client library to httpx2
* Adds per-call API key support for embedding commands
* Enables chaining of configuration and prompt templates
