fabricatio_novel.actions.novel_rag

Novel RAG actions for retrieval-augmented generation.

Classes

GenerateChaptersFromScriptsWithRAG	Generate chapter contents from scripts with RAG writing style augmentation.
StoreWritingStyleTexts	Store writing style reference texts from files into LanceDB.

Module Contents

class fabricatio_novel.actions.novel_rag.GenerateChaptersFromScriptsWithRAG(/, **data: Any)

Bases: fabricatio_novel.capabilities.novel_rag.NovelComposeRAG, fabricatio_core.Action

Generate chapter contents from scripts with RAG writing style augmentation.

Mirrors GenerateChaptersFromScripts but inherits NovelComposeRAG so that create_chapters() applies script-level and scene-level style injection before generating prose. The writing_style_query is the raw user intent (e.g. “Hemingway terse prose style”) and is used for LanceDB retrieval — enable use_refined_query on the fetch config (or the WritingStyleFetchConfig instance) to expand it into multiple semantically diverse queries via the LLM before search.

novel_draft: fabricatio_novel.models.draft.NovelDraft | None = None

The novel draft (for language, metadata).

novel_scripts: List[fabricatio_novel.models.scripting.Script] | None = None

The list of chapter scripts to expand into full text.

novel_characters: List[fabricatio_character.models.character.CharacterCard] | None = None

The list of characters to provide context.

chapter_guidance: str | None = None

Guidance for writing chapter.

writing_style_query: str | None = None

Raw user writing-style intent. Used for both LanceDB retrieval (as the search query) and, when use_refined_query is enabled on the fetch config, as the input to LLM-driven query refinement.

writing_style_fetch_config: fabricatio_novel.models.novel_rag.WritingStyleFetchConfig | None = None

Optional fetch configuration override. Set use_refined_query=True on it to enable LLM-based query refinement; set refined_query_count to control how many variants are produced.

use_reranker: bool = False

When True, embedding search fetches limit * rerank_scale_factor docs, then reranks to limit.

use_refined_query: bool | None = None

Convenience override for WritingStyleFetchConfig.use_refined_query. Has no effect when writing_style_fetch_config is supplied explicitly (use the config field instead). Tri-state: None leaves the default (False); True/False toggles refinement on/off.

refined_query_count: int | None = None

Convenience override for WritingStyleFetchConfig.refined_query_count. Ignored when writing_style_fetch_config is supplied explicitly.

refine_query_template: str | None = None

Convenience override for WritingStyleFetchConfig.refine_query_template. Ignored when writing_style_fetch_config is supplied explicitly.

output_key: str = 'novel_chapter_contents'

Key under which the generated chapter contents will be stored in context.

ctx_override: ClassVar[bool] = True

Whether to override the instance attr by the context variable.

class fabricatio_novel.actions.novel_rag.StoreWritingStyleTexts(/, **data: Any)

Bases: fabricatio_rag.actions.db.StoreTextFile, fabricatio_novel.capabilities.novel_rag.NovelComposeRAG

Store writing style reference texts from files into LanceDB.

Reads text files, splits them into chunks, and indexes them in the LanceDB vector store for later RAG retrieval. This is a standalone ingestion workflow — it does not generate novel content.

store_model: Type[fabricatio_novel.models.novel_rag.WritingStyleDocument]

output_key: str = 'stored_count'

The key used to store this action’s output in the context dictionary.

store_config: fabricatio_lancedb.capabilities.lancedb.LancedbAddRAGConfig | None = None