Planned Release Date: Q4 2026

About This Software

A novel editor with a built-in agent, for writing a novel that is entirely your own.

You must provide your own third-party API key. All API usage fees are settled directly between you and the corresponding AI provider; this software does not collect fees on their behalf.

Please try the Demo version first to see whether the editing workflow fits your needs.

No AI can do this: type one sentence and get tens of thousands of words of high-quality novel.

The novel must be broken down into design steps so the agent can complete the task step by step and refine the novel along the way. In between, a human needs to guide the general direction of the story and clearly state the writing style you expect.

A whole novel can be roughly divided into: story background, character design, chapter outline, and manuscript. (See the software's help pages.)

The output of each step should ideally be no more than 3 times the input length; otherwise text quality tends to degrade.

Directory Structure

A novel is made up of a large number of documents, arranged by number and version.

Multiple versions of the manuscript can be kept independently so you can compare the differences between versions.

Each document's name, tag, color, and role can be edited.

Document tags can be changed in the novel configuration. The defaults include: background design, characters, outline, manuscript, volume 1, volume 2.

Role is an advanced setting used to adjust and strengthen the AI's attention. (By default, just use User for everything.)

The agent selectively reads parts of the novel based on document tags, which reduces token consumption.

The "Big Fat Whale" Agent

A built-in agent that can read your chapters and settings, search the full text, edit the manuscript, and save a revision as a new version.

It tracks task progress on a to-do board and asks you questions when it is unsure.

Before it actually starts writing a new chapter, it submits a proposal and only proceeds once you approve it.

Supports streaming output (Stream), multiple levels of reasoning effort, and context compression.

Built-in background design, character design, and outline templates; the agent guides you through generating the corresponding novel settings.

Task Generation Mode

Directly control the input documents, prompts, and other information of AI calls; includes batch generation mode, chained task mode, document merging, and more.

It is fairly complex; see the software's help pages for more.

This mode saves more tokens, but using the agent is recommended to lower the learning curve.

Estimated Cost

Effective output is counted as the word count of the final version of the manuscript, and total word count is the sum of all documents.

The estimate uses DeepSeek V4 Flash (reasoning_effort: high) at DeepSeek's official off-peak API prices at the time of writing. Provider prices may change; please refer to your provider's bill.

Agent mode: use the agent to assist in generating the novel. As the total word count grows, the agent tries to read in more "possibly related" character information and chapters, which drives the cost up.

Task generation mode: no agent at all; you directly control AI input and output, generating the outline, first draft, and final draft in turn.

• Agent mode, total 0–50,000 words: $1 per 18,000 words of effective output.

• Agent mode, total 50,000–200,000 words: $1 per 15,000 words of effective output.

• Agent mode, total 200,000+ words: $1 per 12,000 words of effective output.

• Task generation mode, 6,000–8,000 words per chapter: $1 per 200,000 words of effective output.

• Task generation mode, 10,000 words per chapter: $1 per 300,000 words of effective output.

Read-Only Mode

After generating a novel, you can access it from a PC or phone over the local network.

With the default configuration, the process only allows access from the local machine; local network access requires changing the configuration. See the help pages for more.

Filter specific chapters by document tag and maximum version number.

Compare text differences between different versions of the same chapter.

Safety Guardrails

The agent refuses to generate obscene or pornographic, illegal, or rule-breaking content.

Different models may have different safety policies and false-refusal behavior. If you are refused by mistake, switch to another model and try again.

Demo Version Limits

The Demo version keeps the core features but limits usage; the full version removes the limits.

• Up to 2 novels

• Per novel: up to 25 manually created documents, up to 50 agent-created documents

• Parallel tasks: up to 1

• Agent sessions: up to 2

App Configuration

Launch the app in "Configuration Mode" to set up your third-party API keys.

Then close the app and launch it in "Application Mode".

AI Providers

Supported AI providers: DeepSeek, OpenRouter, Volcengine, TokenMix.

The software is optimized for the DeepSeek API first; other models are only compatibility-tested.

Some models with fewer parameters or lower capability do not support tool calling (Tools); larger models are recommended.

Third-party trademark notice: DeepSeek, OpenRouter, Volcengine, and TokenMix belong to their respective owners.

For Windows users, API keys are stored in a local file and protected with DPAPI encryption.

Other platforms are not supported yet.

Novel text and your input are sent to the AI provider in order to call the AI service.

AI Generated Content Disclosure

The developers describe how their game uses AI Generated Content like this:

The software uses AI images.
When using the software, users need to provide an API key to call third-party AI services.
The AI service costs you use are paid with your own API key.