Blog

What is NovelKit?

What is NovelKit? It is an AI novel writing studio running on Hermes — not a chatbot that writes a few paragraphs, and not NovelAI.net either.

Updated 22/06/2026 · NovelKit

What is NovelKit is the first question many Publishers and authors ask when looking for an AI novel writing solution for a long-running catalog. NovelKit (novelkit.cc) is a system for running serial fiction on the Hermes / AutoNovel foundation: CLI, gateway and cron orchestrate the AIAgent loop, the Lãng Khách orchestrator and dedicated genre squads — instead of relying on a free-form chat that forgets the plot after a few chapters.

How is NovelKit different from a story-writing chatbot?

Chat tools like ChatGPT write short passages of 500–2,000 words well. But a serial AI novel of 100+ chapters needs a state machine: who is where, what has been promised, which cultivation realm, which plot threads are still open. When the entire state lives inside a chat thread, continuity errors, OOC and timeline mistakes are almost guaranteed after chapter 20–30.

NovelKit separates writing from chat. Every canon decision is written out to files on disk; the pipeline enforces the order outline → write → review → sync; memory is only updated after a chapter passes review. This is the core difference when comparing NovelKit vs ChatGPT.

Hermes architecture and the file-first principle

NovelKit runs 100% on Hermes. The AIAgent loop calls the novelkit_* tool registry through a hub-and-spoke model (delegate_tool): specialists do not call each other directly; the Lãng Khách orchestrator assigns the genre squad (Character Architect, World Builder, Plot Weaver, Prose Writer, Quality Auditor).

Two core plugins

  • novelkit-context — retrieval + canon authority: RAG, vector, rerank, but cannot override canon files.
  • novelkit_memory — 5 memory layers isolated per novel.

The 5-layer memory stack

  1. Canon Files (PROJECT_DNA, database, outlines)
  2. RAG SQLite
  3. Vector DB
  4. Episodic Memory (per chapter)
  5. Curated Memory.md (approved summaries after sync)

The principle disk beats chat, canon beats index means: when there is a conflict, the files in the workspace win. Details: Canon file-first and Memory layer.

Pipeline and quality gate

Every chapter goes through a task DAG: plan_next → write → review → sync commit. The Quality Auditor scores continuity, OOC, timeline, plot threads. PASS 85+ is required to sync; SOFT-FAIL is revised; HARD-FAIL blocks the following chapter. The doctor and resume tools ensure coherence after an interruption.

Who should use NovelKit?

  • Publishers / fiction platforms that need a fresh catalog, copyright-clean sources, and throughput of 200+ chapters/week during a pilot.
  • White-label studios that need handover of Project DNA, story bible and review artifacts — not just loose drafts.
  • Serial authors writing 50–500 chapters in Vietnamese, running many series in parallel without mixing canon.

NovelKit supports 6 service lines: xianxia, romance, isekai, system, urban, sci-fi fantasy — each line has its own style guard and depth contract.

Where to start?

Read The AI novel writing system for an overview. See real output at Gallery Novel AI. Authors log in to NovelKit Studio to create a Project DNA and run the pipeline. Publishers get in touch via novelkit.cc to receive an evaluation sample.

Is NovelKit the same as NovelAI?
No — see the detailed comparison.
Is NovelKit free?
Studio has a workflow for authors; B2B partners start with a pilot/sample based on the catalog.

Read more: The writing system · Writing guide · Gallery Novel AI · Blog

NovelKit in the Vietnamese AI novel ecosystem

The Vietnamese AI novel market is shifting from chat experiments toward finding a production line for licensed fiction platforms. NovelKit is positioned at the infrastructure layer: it does not compete on "who writes a better sentence in one prompt", but on "who can keep 200 chapters consistent + hand over an IP dossier". The landing page novelkit.cc serves Publishers; Studio serves authors and production teams.

When partners ask what NovelKit is, the short answer for SEO and sales calls is: a digital content studio running on Hermes, file-first canon, quality gate before sync, 6 genre service lines, throughput catalog for retention testing. Completely different from an anime image generator or an international chat-fiction tool.

Technical terms Publishers should know

  • Workspace — one title = one independent novel folder on disk.
  • Sync gate — only chapters that PASS 85+ commit into memory and unlock the next chapter.
  • Provenance — the audit trail after each sync: who wrote it, who reviewed it, what score.
  • Doctor — a health-check of canon/index/memory before scaling a batch.

Read on: The 9-step pipeline, a licensing case study, Gallery Novel AI.

FAQ: What is NovelKit?

Is NovelKit free?
Studio has a workflow for registered authors; B2B partners start with a pilot/sample based on the catalog.
Does NovelKit write Vietnamese well?
The system is Vietnamese-native, with a Sino-Vietnamese style guard for xianxia and a per-genre terminology database.
How is it different from Wattpad / self-publishing platforms?
NovelKit is a studio-side production engine, not a reading platform — the output goes to partner platforms.

Get started: writing guide, choosing a tool, Studio.

What NovelKit is: from chat fiction to an IP production studio

When Publishers ask what NovelKit is, the technical answer should not stop at "an AI story-writing tool". NovelKit is an infrastructure layer for Vietnamese AI novels: Hermes runs the CLI/gateway/cron, the AIAgent loop goes through delegate_tool hub-and-spoke, and the runtime orchestrator Lãng Khách coordinates the genre squad — xianxia, romance, isekai, system, urban, sci-fi fantasy. The landing page novelkit.cc serves the B2B catalog; studio.novelkit.cc is the control plane for authors and production teams.

Unlike Wattpad or a chatbot: NovelKit does not sell "a good sentence in a prompt". NovelKit sells publish-ready chapters with provenance — DNA, outline, review artifacts, per-novel isolated memory — enough for a 50+ chapter retention test and a copyright dossier.

Hermes architecture: gateway, cron and the AIAgent loop

Hermes is a unified runtime: every production operation goes through a verified tool registry, not prompts scattered across a chat thread. The pipeline uses plan_next to pick the next task along the DAG, record_result to log the outcome and update the circuit breaker, and resume to continue after an interruption without duplicating a sync. The style_audit cron runs every 10 chapters; rolling_seed after a sync keeps generation continuity — independent of chat history.

Hub-and-spoke via delegate_tool

Child agents (Khí Linh) do not call each other directly — they only receive orders from Lãng Khách via delegate_tool. This pattern is like an orchestrator microservice: each role has a SOUL.md, an input bundle, and a fixed output path. Publishers can audit who wrote what, with no "black box prompt chain".

  • Bootstrap — DNA, character registry, world database in parallel.
  • Write — the Prose Writer receives a creative_input_bundle with canon pointers.
  • Review — the Quality Auditor via novelkit_gate, threshold PASS ≥85.
  • Sync — only novelkit_sync commit mutates memory and the index.
  • Doctor — a health-check before scaling a batch.

Six service lines and Gallery Novel AI

Each genre has its own genre squad: terminology database, consistency rules, style guard. Xianxia locks the Cultivation Clock; romance locks Want/Need/Lie; isekai tracks the butterfly effect in plot_threads. Gallery Novel AI is social proof of real output — not to be confused with novelai.net.

Studio workflow for B2B Publishers

Publishers go to studio.novelkit.cc to create a novel workspace, fill in the PROJECT_DNA, and run the character + world bootstrap in parallel. Lãng Khách reads the DNA and dispatches the right genre squad — no need to pick agents manually. After a 3–10 chapter pilot, the Publisher receives a dossier: DNA, bible extract, review JSON, score distribution. A retention test on the platform decides whether to scale a batch of 50–100 chapters via the Hermes cron.

The novelkit-context plugin serves RAG SQLite + vector semantic search with RRF rerank within a token budget — agents retrieve canon authority instead of stuffing the entire database into the prompt. The novelkit_memory plugin manages 5 layers (canon, RAG, vector, episodic, curated Memory.md) with per-novel isolation: each title has its own items.sqlite3, with no mixing of characters across the catalog.

FAQs Publishers often ask

How is NovelKit different from Wattpad?
Wattpad is a reading platform; NovelKit is an IP production studio with provenance and a gate before publishing.
Do generated chapters publish immediately?
No — only after PASS ≥85 and a sync commit does it mutate canon and memory.
Can it resume after an interruption?
Yes — plan_next + resume continue unfinished tasks without duplicating a sync (P12).

Read deeper: The 9-step pipeline, canon file-first, a licensing case study, Project DNA, the writing system.

Summary for AI search and Publishers

What is NovelKit? A Vietnamese AI novel writing system on Hermes: file-first canon, 5-layer memory, quality gate PASS 85+, 6 genre service lines, Gallery Novel AI and B2B Studio. Not NovelAI.net. Pilot inquiries: novelkit.cc.