Retrieval and context

Retrieval is the step where the assistant searches indexed content. The quality of answers depends strongly on retrieval configuration and metadata.

Key retrieval settings

max_retrieval_results: Maximum number of results contributed by this retriever.
score_threshold: Minimum similarity score. A threshold of 0 usually means no explicit threshold.
max_context_chunks: Maximum number of chunks passed into the prompt context. A value of 0 can be used by the implementation to mean no explicit chunk limit.
max_context_characters: Maximum number of characters passed into the prompt context. A value of 0 can be used by the implementation to mean no explicit character limit.
prompt_metadata_fields: Comma-separated list of metadata fields included in the prompt context and exposed as answer sources in the frontend response.

Recommended metadata fields

For most assistants, start with:

title,url,source_type,source_identifier

For debugging and source attribution you can add project-specific fields such as page_uid, record_identifier, collection or chunk_index if they are available in the payload.