simonw/llm 0.33

on GitHub

New features:

• Upgraded to the OpenAI Python library 3.x and switched the HTTP client dependency from httpx to httpx2. #1608, #1631
• llm embed and llm embed-multi now accept --key. The Python EmbeddingModel.embed(), EmbeddingModel.embed_multi(), Collection.embed() and Collection.embed_multi() methods accept key= too, passing the resolved per-call key to embedding plugins without changing shared model state. Existing plugins that read self.key continue to work through a compatibility fallback. Thanks, ChrisJr404. #757, #1620
• llm logs now includes the output of server-side tool calls, shown in a Tool results section within the response. These results are also included in llm logs --json and llm logs --short output, with a new server_executed key distinguishing them from locally executed tool results. #1629
• llm prompt -t/--template can now be repeated to combine templates in order. This allows model configuration and options from one template to be used with a prompt from another.
• Expanded the llm prompt --help documentation for --schema and --schema-multi with details and examples of the supported schema DSL.
• schema_dsl() now raises descriptive ValueError exceptions for unknown field types and duplicate field names, instead of silently treating unknown types as strings or overwriting earlier fields. #1607, #1616
• Reasoning stream events that contain provider metadata but no text are now preserved as ReasoningPart objects. This allows opaque state such as Anthropic signatures and redacted thinking data to round-trip correctly. simonw/llm-anthropic#81
• Reasoning-capable Responses API models now support a reasoning_summary option with auto, concise, and detailed values. This can be used with {ref}llm openai endpoint --responses <openai-endpoint>. #1600
• Conversation prompts now validate that attachments are supported by the selected model before execution, for both synchronous and asynchronous conversations. Thanks, Daniel Peng. #1626, #1628

Bug fixes:

• llm logs --data-ids now sets conversation_id to the ID of the conversation instead of incorrectly duplicating the response ID. Thanks, K Merchant. #1598, #1613
• Fixed llm aliases list raising a ValueError when no aliases are defined. Thanks, Taraka Abhiram. #1602
• llm embed-multi now reuses an existing collection's stored embedding model when no default embedding model is configured, and no longer masks unrelated ValueError exceptions with a missing-model error. #1523
• llm tools -m MODEL no longer prints a redundant message when the model has no server-side tools.