InkoS: a multi-intelligence system for automated generation and proofreading of long stories

InkoS (Ink Operating System) is a multi-agent automated writing system for long novels, open-sourced by developer Narcooo in 2026. Faced with the problems of character amnesia, broken strokes, contradictory logic, and vocabulary fatigue that often occur when traditional AI generates long texts, InkoS adopts a rigorous engineering approach and creates a unique “Truth Files” underlying architecture. The system accurately deconstructs the creative process into three distinctly functioning intelligences: Writer, Auditor and Revisor, which align and update the seven categories of plot knowledge and worldview status in real time during the creative process. The system not only supports fully-automated assembly-line continuous creation from scratch, but also has strong “reverse engineering” capabilities: it allows users to import the text of existing unfinished novels, and the system will automatically parse and reconstruct the worldview to realize a completely seamless succession of writing. With its flexible endpoint-based design, InkoS provides fully automated mode, atomic JSON command mode, and natural language interaction agent mode, making it an industrial-grade infrastructure for professional text creators and AI application developers to control high-quality long texts.

Function List

• Truth Files Underlying ArchitectureThe book's main objective is to eliminate the AI hallucinatory disasters generated by long, multi-chapter books by dynamically maintaining seven separate sets of truth files for each book, tracking character knowledge, resource depletion, worldview constraints, and vocabulary usage frequency in real time.
• Reverse engineering and automatic renewal of existing works: Support for importing local single-file novels or chapter directories (support by regular expression) --split (Customized sub-chapters and breakpoint continuation guide). The system can automatically disassemble the old text and reverse generate a complete character matrix and ambiguous state, and then turn on continuous seamless continuation with a single click.
• Fully automated multi-intelligent body content closure: Separate job pipelines. The drafter intelligence focuses on vivid descriptions; the auditor intelligence examines plot breaks according to dozens of checking dimensions; and the revisionist intelligence directly fixes the text in response to the review report.
• Human Review Gates (HRG) mechanism: It supports the setting of multi-chapter batch continuous operation, automatically hangs the task after generating a specified number of chapters, and forcibly introduces human supervision and gatekeeping to ensure that the main direction of the story is in line with the creator's original intention.
• Full coverage of the three operational dimensions: Support for “one-click complete pipeline” generation; support for “atomized separation commands” with fine-grained control (via the --json (interfacing output for access to external workflows); supports a “Natural Language Agent” session mode with 13 separate tools built in.
• Automated vocabulary fatigue detectionAI: Get rid of the heavy “machine flip” flavor of AI. The system comes with a vocabulary monitor that avoids high-frequency repetitive adjectives and clichés, maintaining a sense of authenticity and richness in the tone of the text.

Using Help

1. Basic environment preparation and system installation

InkoS is a powerful TypeScript command-line application (CLI), and before running it, you need to make sure that your computer has a standard development runtime environment.

• Configuring the underlying environment: Please go to the official Node.js website to download and install the latest LTS version (Node.js 18.x and above is recommended).
• Download and Global Installation：
  Open your system's terminal program (PowerShell for Windows or Terminal for Mac/Linux), clone the code repository, and finish compiling and linking:

  git clone https://github.com/Narcooo/inkos.git
  cd inkos
  npm install
  npm run build
  npm link
  

• Configuring the Big Model Key (API Key)：
  The powerful self-censorship and text generation capabilities of InkoS rely on the Top Big Language model. Please set your credentials in the current terminal environment variable:

  export OPENAI_API_KEY="您的_API_KEY_填写于此"
  

2. Core workflows in detail: three modes of operation

InkoS is designed to be used in three modes of operation, so that you can find the best way to use it, whether you are “inspired and ready to go” or “need to polish every sentence”.

Mode 1: Complete pipeline mode (one-click fully automated production)

This model is the most productive for situations that require quick turnaround, as InkoS schedules “writers,” “auditors,” and “revisers” in sequence and delivers the final draft directly.

• Single Chapter Generation and Refinement：

  inkos write next 吞天魔帝
  

  Upon input, the system fully automatically reads the settings and writes the next draft chapter; a multi-dimensional logic audit is then initiated (to ensure that the character is not using non-existent props); and finally the revision module polishes and finalizes the output.

• Batch Hanging Creation Mechanism：

  inkos write next 吞天魔帝 --count 5
  

  The system will advance through the 5 chapters in a continuous and unobtrusive manner, fully automatically. There is no need for any human intervention at this point, and when it is finished, it is automatically deposited in the catalog for your review.

Mode 2: Atomic Command Mode (Fine Control and Third Party Integration)

The model is intended for advanced users or developers, and can be forced to dismantle the entire closed-loop pipeline and inject personal will into each step of the process.

• Restricted outline drafting：

  inkos draft 吞天魔帝 --context "本章重点写师徒矛盾爆发，主角愤而离家出走" --json
  

  --context Parameters provide you with entrances to forcibly intervene in the direction of the plot; and the --json The output is fully structured, making it extremely easy to automate scheduling in conjunction with external agents such as OpenClaw.

• Independent call to the audit module to validate individual manuscripts：
  If you've hand-written the latest chapter yourself, you can use the following commands to have the AI act as a proofreader for you, picking out logical holes:

  inkos audit 吞天魔帝 31 --json
  

• Automated symptomatic implementation of changes：

  inkos revise 吞天魔帝 31 --json
  

Model 3: Natural Language Intelligentsia (Agent) Model

Don't want to memorize complicated commands? You can direct InkoS to work directly in natural language.
Up to 13 powerful tools are loaded inside the system (e.g., creating books). create_bookRead the truth papers. read_truth_filesScanning the market scan_market etc.), it will automatically plan thought links.

• Situational dialogues to give instructions：

  inkos agent "帮我写一本都市修仙类型的小说，主角是个程序员，马上帮我建档并写好第一章。"
  

• Composite Level Strategy Arrangement：

  inkos agent "先扫描目前的网文市场趋势，然后根据热门数据为我创建一本新书的大纲。"
  

3. Advanced features: import of original works and seamless continuation of writing at breakpoints

If you've got a million-word novel in your hands that's been abandoned or is currently being serialized, and you want to use AI to take the stress out of writing it, InkoS's “reverse engineering” is perfect for taking over the thinking.

• Step A: Loading the Original Novel
  Save the text of your written chapter as a local file.
• Step B: Initiate reverse disassembly import

  inkos import 吞天魔帝 --file ./novel.txt --split "第[一二三四五六七八九十百千万\d]+章"
  

  included among these --split Parameters support the input of regular expressions to flexibly adapt to the requirements of various typesetting formats of subchapters. If the import is too long and pauses in the middle, you can always use the --resume-from <章节号> Breakpoint Recovery.

• Step C: Worldview Reconstruction and Truth Document Generation
  Instead of manually organizing your data, the system will automatically read through your text during import, reverse combing and generating all 7 truths (including character status, unresolved ambiguities, world limitations, etc.).
• Step D: Fully automated continuation
  fulfillment inkos write next 吞天魔帝The system will follow your original style of writing and continue the story for you in a logical manner. The system will strictly follow your original style of writing, inherit all the previous strokes, and logically continue to write the story for you.

application scenario

1. High Frequency Updating and Capacity Expansion of Online Long Story Serialization
   Faced with the burden of 10,000-word updates per day, web authors are often stuck in a rut, or their energy fails, resulting in a logical breakdown of the plot. InkoS's fully automated writing, cross-auditing, and auto-revision pipeline quickly produces a draft based on a one-sentence outline assigned by the author and self-eliminates the obvious poisonous points, so that the creator can retain his core energy in the outline planning, and geometrically multiply the production capacity of the writing process.
2. Resurrection of an old pit that was discontinued and seamless continuation of an old work to fill in the hole
   Creators who want to fill in the gaps of years gone by or pick up where they left off. With InkoS's unique import command, you can feed the system with your past manuscripts. The system can then reverse-engineer the characters and unanswered questions of the old work and take over the next storyline in a highly adaptable style.
3. Complex game plot combing and dynamic set maintenance
   Large-scale narrative videogames and tabletop running games (TRPGs) have extremely complex character dynamics and resource flows, and the InkoS kernel comes with a highly structured “truth file” that tracks the decay and change of various worldview states as accurately as database changes, making it an ideal management tool for dynamic event setups in games, in addition to for writing. In addition to being used for writing, it is also an ideal tool for managing a game's dynamic event set.
4. Underlying Infrastructure for Third Party Intelligentsia and Secondary Developers
   For developers who aspire to build advanced shell AI products or automated creation pipelines, InkoS provides JSON atomized commands that can be directly called by external agents as underlying API components. The high degree of assembly freedom allows it to seamlessly interface with various scheduling frameworks, making it easy to build unattended content farms or commercial authoring platforms.

QA

1. InkoS How do you avoid the problem of AI “forgetting” previous settings or “making up” props in long creations?
   The system does not rely at all on the fragile contextual history of large language models. It has a unique “Truth Files” engineering architecture. The system creates seven separate local truth state files for each book to record worldview constraints, character attribute increases and decreases, and resource consumption. Every time before writing or auditing, AI must and can only use the above Truth Files as the only objective basis for the current events, which eliminates the illusion of a long book from the mechanism.
2. Can I just have InkoS review my own manuscript logic for me?
   InkoS supports instruction-level decoupling. You don't need it to generate text, you can just use the inkos audit <书名> <章节数> Atomic commands call the Audit Intelligence alone, which automatically compares the underlying setting state of the entire book and outputs a report on the plot logic breaks and setting gaps in your handwritten manuscript.
3. What about feeding it hundreds of thousands of words of my own novel that I've written and what about the messed up sub-chapter layout?
   InkoS provides extremely flexible importing function for old novels, and the single-file import mode automatically slices and dices the novels according to the common “Chapter X” by default. If your novel has a special layout, the system supports introducing regular expressions (e.g. --split <regex>) for precise slicing and dicing. Even if the power failure interrupts in the middle of mass import, the progress can be quickly resumed by utilizing the breakpoint continuation parameter.