Retrieve Metadata from Assistant Replies

Retrieves metadata from assistant replies within an LLMMessage object. It returns the metadata as a tibble.

Usage

get_metadata(.llm, .index = NULL)

last_metadata(.llm)

.llm: An LLMMessage object containing the message history.
.index: A positive integer specifying which assistant reply's metadata to extract. If NULL (default), metadata for all replies is returned.

A tibble containing metadata for the specified assistant reply or all replies.

Metadata columns may include:

model: The model used for generating the reply.
timestamp: The time when the reply was generated.
prompt_tokens: The number of tokens in the input prompt.
completion_tokens: The number of tokens in the assistant's reply.
total_tokens: The total number of tokens (prompt + completion).
cached_tokens: Prompt tokens served from the provider's prompt cache. NA_integer_ for providers that do not report cache usage; 0 means a reported cache miss.
cache_creation_tokens: Prompt tokens written to the cache. Currently reported by claude() only; NA_integer_ elsewhere.
api_specific: A list column with API-specific metadata, including the raw provider-specific cache fields.

For convenience, last_metadata() is provided to retrieve the metadata for the last message.