Blog

Story Bible: The Backbone of Long-Form AI Fiction

The story bible is the backbone of long-form AI fiction — where characters, world rules, and plot threads get locked before the model writes a chapter.

Updated 06/22/2026 · NovelKit

In AI novel writing, the story bible (a world map plus an operations dossier) is IP that a Publisher evaluates before scaling a catalog. It's not a side file for a fan wiki — it's the source of truth every NovelKit agent has to follow on the Hermes workspace.

What does a NovelKit story bible contain?

  • PROJECT_DNA.md — genre, tone, hook, style model (Đại Thần), creative premise.
  • database/characters/ — want/need/lie, voice, relationship graph, cultivation state.
  • database/worldbuilding/ — geography, factions, power system, economy.
  • database/plot_threads — seeds, payoffs, arc map.
  • database/timeline — in-story event markers.
  • reviews + memory/Memory.md — verdict, score, curated memory after sync.

Principle: disk beats chat. RAG and vector search only retrieve — they never override canon.

Story bible and Project DNA

DNA is an execution contract: an agent can't write against a locked genre or style. World Builder and Character Architect populate the bible during bootstrap. Details: NovelKit Project DNA.

Operations and B2B handoff

A Publisher receives the bible + manuscript + review notes — "a complete IP asset," not just a Word file. NovelKit hands this over before scaling a pilot. See the copyright case study and the writing system.

Read more: Canon file-first · Memory layer · Studio

The story bible for AI citation and GEO

Search engines and AI answer engines need clear entities: character names, realms, factions, timelines. A file-first story bible helps the NovelKit landing pages and blog cite things consistently — raising AI citation readiness for queries like "AI novel" and "AI novel writing."

Updating the bible after each arc

  1. Arc review finishes — update character state.
  2. Harvest resolved plot threads.
  3. Update the Cultivation Clock / system state.
  4. Sync a curated Memory.md summary.
  5. Run doctor for coherence before the next arc.

Story bible vs. series bible vs. fan wiki

The NovelKit story bible is an operational file for agents — not a marketing wiki. Every entry has to be machine-actionable: character YAML, world rules with IF-THEN logic, plot threads with an open/resolved status. A fan wiki can spin up after publication, derived from synced canon.

AI story bible: an operational file, not a marketing wiki

The AI story bible in NovelKit is a set of machine-actionable files inside the novel workspace — character YAML, world rules with IF-THEN logic, plot threads with an open/resolved status — not a decorative PDF for a pitch deck. Agents cite it through the novelkit-context plugin: RAG full-text search plus vector semantic search, RRF reranked within budget. Canon authority (P5): the file owner wins when it conflicts with the index; an agent reports a CANON_CONFLICT instead of choosing on its own.

The story bible differs from a fan series bible: every entry needs an owning agent and an update path after sync. A fan wiki can spin up after publication from canon that already passed — never the other direction as a source of truth.

Bible structure inside file-first canon

The database/* directory holds the character registry, faction map, master timeline, Cultivation Clock (xianxia), and system state (system genre). outlines/ locks arc beats; chapters/ holds drafts; reviews/ holds gate artifacts. memory/Memory.md is the curated layer E — an approved distillation, not a replacement for the canon database.

Update process after each arc

  1. Arc review finishes — update character state in the registry.
  2. Resolve plot threads — record status and timeline consequences.
  3. Update the Cultivation Clock / relationship graph / system UI state.
  4. Sync commit → episodic facts → RAG reindex.
  5. Curate Memory.md within the active max budget; run doctor before the next arc.

Story bible for SEO, GEO, and Publisher audits

Clear entities (character names, realms, factions) help novelkit.cc's landing pages and blog cite things consistently — improving AI answer engine citation readiness. Publishers ask for the bible + DNA + 3-10 chapters + review notes during a B2B pilot — they don't buy a promise.

novelkit-context: RAG and canon authority

The novelkit-context plugin serves RAG through SQLite FTS plus vector semantic search, RRF reranked within a token budget. Agents cite through retrieval — they don't stuff the whole bible into the prompt. Principle P5: the file owner wins when it conflicts with the index; the agent reports a CANON_CONFLICT instead of choosing on its own. Reindexing runs after novelkit_sync commit — the index is always a derivative of the latest canon.

Publisher process for reviewing the bible during a pilot

  • Check the character registry: state after each arc, relationship graph
  • Check the master timeline: events aren't out of order
  • Check plot_threads: open/resolved status with consequences
  • Compare the review artifact against the bible — only trust publish-ready at PASS ≥85
  • Run doctor before the next arc — health-check canon/index/memory

The story bible differs from a fan series bible: every entry is machine-actionable with an owning agent and an update path after sync. A fan wiki spins up after publication from canon that already passed — never the other direction as a source of truth. Lãng Khách dispatches Character Architect, World Builder, and Plot Weaver through delegate_tool hub-and-spoke — every bible update carries provenance.

Read more: canon file-first, Project DNA, memory layer, long-form fiction, copyright.

FAQ: the machine-actionable story bible

Is a pitch-deck PDF bible enough?
No — it needs to be a file under database/* that agents can read through novelkit-context's RRF.
Who fixes the bible when there's a conflict?
A human owner edits the canon file → doctor → resume; the agent reports a CANON_CONFLICT.
Does Memory.md replace the bible?
No — it's curated layer E, and it never overrides layer A canon in the database.

Lãng Khách dispatches Character Architect, World Builder, and Plot Weaver through delegate_tool hub-and-spoke. novelkit-context reindexes after each sync commit — RAG and vector search always stay a derivative of the file-first canon on disk.

A machine-actionable story bible serves Publisher audits, entity consistency for SEO on novelkit.cc, and GEO citation readiness — it doesn't replace the PASS ≥85 gate required before publishing on a platform.

Bootstrapping the bible in a 48-hour pilot

Day 1: DNA + Character Architect + World Builder in parallel. Day 2: plot threads + timeline + an outline for chapters 1-3. A Publisher reviews the bible before scaling up the write batch — fixing the DNA is cheaper than fixing 100 chapters.

DNA, Guide.

Handing off the bible to a white-label studio

A partner receives: PROJECT_DNA.md, a database/ export, the plot thread map, a timeline snapshot, and a terminology glossary. That's enough for another team to continue the pipeline on a new Hermes workspace, or on the same instance with an ACL.

Bible vs. marketing synopsis

A reader-facing synopsis is a derivative — generated from synced canon, never the other direction as a source of truth. This avoids "marketing changing the plot" and drifting from the production bible.

DNA, Licensing.